JP2020119199A - Load estimation device and load estimation method - Google Patents

Abstract

To provide a load estimation device that easily obtains partial load information of an estimation object user without installing a measurement device for measuring a partial load in the estimation object user and accurately estimates the partial load information of the estimation object user.SOLUTION: Load information and a feature amount of a user are classified, according to similarity between partial load ratios, into a primary cluster. The load information belonging to each primary cluster is classified, according to similarity between total load information, into a secondary cluster. The load information belonging to each secondary cluster is classified, according to similarity between partial load information, into a tertiary cluster. A representative model representing the load information belonging to each tertiary cluster is created. A selected representative model is selected, when total load information and a feature amount of an estimation object user are given, from representative models that represent load information belonging to a selected primary cluster that is derived by a derivation formula. Partial load information included in the selected representative model is reflected in partial load information of the estimation object user.SELECTED DRAWING: Figure 3

Description

The present invention relates to a load estimation device and a load estimation method for estimating partial load information indicating a temporal change of a plurality of partial loads each indicating a magnitude of electric energy consumed by a plurality of portions of an estimation target consumer.

With the spread of smart meters in recent years, a system for measuring the total load indicating the total amount of electric energy consumed by a consumer every 30 minutes and collecting the measured total load is becoming widespread. Therefore, it is easy to understand the total load.

On the other hand, when it is possible to understand not only the total load but also the load for each device that indicates the amount of electrical energy consumed by multiple devices used by consumers, the power supply source The accuracy of control can be improved, the consumer can estimate the outcome of the action for energy saving, and the profit is generated for both the power supply source and the consumer.

The load by device may be found by directly measuring the electric energy consumed by the device, or by estimating the load by device from the total load.

In the technique described in Patent Document 1, five measuring instruments each obtain measurement information regarding power consumption in an air conditioner, a television, a microwave oven, a refrigerator, and a lighting fixture, and transmit the obtained measurement information to a management system. (Paragraphs 0031 and 0032).

In the technique described in Patent Document 2, a power consumption pattern of the entire device and a typical power consumption pattern of each device are registered in the database (paragraph 0016). In addition, the measuring means records time-series data indicating the time change of the power consumption of the entire device in the time-series database (paragraph 0013). Further, the time-series complementing unit selects a pattern of the entire device that is most similar to the time-series data stored in the time-series database or has a distance within a threshold value from the patterns of the entire device in the database (paragraph 0021). ). In addition, the time series complementing unit extracts a part from the current time to the end time in the selected pattern and complements it after the time series data (paragraph 0024). In addition, the pattern selection unit extracts from the pattern group of each device a pattern whose sum is the most similar to the complementary pattern or whose distance is equal to or less than the threshold value (paragraph 0026).

In the technology described in Patent Document 3, the clustering execution unit instructs the change over time of the individual load of the consumer, which is the energy consumption of each of the N consumers, and provides monitoring house information including total load and load by device. Based on this, N customers are classified into M clusters (paragraphs 0022 and 0024). Further, the representative model data acquisition unit acquires M types of representative model data based on the customer individual load of at least one customer belonging to the same cluster for each of the M types of clusters (paragraph 0025). In addition, the representative model data selection unit sets one piece of data based on the similarity with the estimation target consumer information that indicates the total load that is the total energy consumption of the estimation target consumer among the M types of representative model data. Is selected as the selected representative model data (paragraphs 0027 and 0028). In addition, the breakdown calculation unit correlates the total load instructed in the estimation target consumer information and the selected representative model data in time series to calculate a plurality of device-specific loads that are a plurality of partial loads in the estimation target consumer. (Paragraph 0029).

JP, 2014-112439, A JP, 2011-176984, A JP, 2017-224121, A

The conventional techniques for grasping the load by device represented by Patent Documents 1, 2 and 3 are all expected behaviors in which a measuring device or the like for measuring the load by device of a target consumer must be installed. There is a problem that the load for each device corresponding to each pattern must be prepared in advance. Therefore, the conventional techniques for grasping the load by device represented by Patent Documents 1, 2 and 3 have a problem that the load by device cannot be easily obtained.

In addition, in the conventional technique for grasping the load by device represented by Patent Document 3, the load by device of each of the plurality of consumers is similar to each other even though the total load of the plurality of consumers is similar to each other. If not, there is a problem in that it is not possible to obtain a device-specific load with high accuracy.

These problems also occur when estimating partial loads other than device-specific loads.

The present invention is made to solve these problems. The problem to be solved by the present invention is to easily obtain the partial load information of the estimation target consumer without installing the measuring device for measuring the partial load in the estimation target consumer, and to obtain the partial load information of the estimation target consumer. It is to estimate accurately.

The present invention is directed to a load estimation device.

The load estimating device includes a partial load ratio calculating unit, a first similarity calculating unit, a primary clustering unit, a derivation formula creating unit, a second similarity calculating unit, a secondary clustering unit, a third similarity calculating unit, A third-order clustering unit, a representative model acquisition unit, a first-order cluster selection unit, a fourth similarity calculation unit, a representative model selection unit, and a calculation unit are provided.

The partial load ratio calculation unit calculates a partial load ratio for each piece of load information that is each of the plurality of load information. Each load information includes total load information and partial load information. The total load information indicates the time change of the total load indicating the total amount of electric energy consumed by one consumer. The partial load information indicates a temporal change of a plurality of partial loads indicating the magnitude of electric energy consumed by a plurality of parts of one consumer. Each load information is associated with a feature amount indicating the feature of one consumer. The partial load ratio indicates the ratio of a plurality of partial loads to the total load.

The first similarity calculation unit calculates a first similarity between a plurality of partial load ratios respectively calculated for a plurality of load information.

The primary clustering unit classifies a plurality of load information and a plurality of feature amounts respectively associated with the plurality of load information into a plurality of primary clusters according to the first similarity. As a result, at least one load information belonging to each of the plurality of primary clusters and at least one feature amount associated with each of the at least one load information are specified.

The derivation formula creating unit learns at least one load information and at least one feature amount belonging to each of the plurality of primary clusters, and derives a primary cluster from the given total load information and the given feature amount. Create an expression.

The second similarity calculation unit is configured to detect a first load among a plurality of total load information included in a plurality of load information belonging to each primary cluster which is each of at least one primary cluster included in the plurality of primary clusters. Calculate a similarity of 2.

The secondary clustering unit classifies the plurality of load information belonging to each primary cluster into a plurality of secondary clusters according to the second similarity. Thereby, at least one load information belonging to each of the plurality of secondary clusters is specified.

The third similarity calculation unit is configured to detect a plurality of pieces of partial load information included in a plurality of load information belonging to each secondary cluster, which is each of at least one secondary cluster included in the plurality of secondary clusters. Calculate the similarity of 3.

The tertiary clustering unit classifies the plurality of load information belonging to each secondary cluster into a plurality of tertiary clusters according to the third similarity. Thereby, at least one load information belonging to each tertiary cluster, which is each of the plurality of tertiary clusters, is specified.

The representative model acquisition unit creates a representative model representing at least one load information belonging to each tertiary cluster. As a result, a plurality of representative models are acquired.

The primary cluster selection unit is total load information provided with total load information indicating a temporal change in total load indicating the total amount of electric energy consumed by the estimation target consumer, and determines the characteristics of the estimation target consumer. When the indicated feature amount is the given feature amount, the selected primary cluster derived by the derivation formula is selected from the plurality of primary clusters.

The fourth similarity calculation unit selects, from a plurality of representative models, two or more representative models that represent the load information belonging to the selected primary cluster, and calculates the total amount of electric energy consumed by the estimation target consumer. A fourth similarity is calculated between the total load information indicating the change over time of the total load shown and the total load information included in each of the two or more representative models.

The representative model selection unit is included in the representative model that is a candidate and the total load information that indicates a temporal change in the total load that indicates the total amount of electric energy consumed by the estimation target consumer based on the fourth similarity. A selection representative model is selected from two or more representative models by using a selection criterion in which a candidate representative model is more likely to be selected as the degree of similarity with the total load information is higher.

The calculation unit reflects the partial load information included in the selected representative model so that the partial loads indicating the temporal changes of the plurality of partial loads indicating the magnitudes of the electric energy consumed by the plurality of parts of the estimation target consumer, respectively. Calculate information.

The invention is also directed to a load estimation method.

According to the present invention, partial load information of the estimation target consumer is obtained from the plurality of load information of the consumer and the plurality of feature amounts, and the total load information and the feature amount of the estimation target consumer. It is possible to easily obtain the partial load information of the estimation target consumer without installing a measuring device for measuring the partial load.

Further, according to the present invention, the total load information similar to the total load information of the estimation target consumer is calculated from the two or more representative models representing the load information of the consumer having the feature amount similar to the feature amount of the estimation target consumer. A selected representative model having load information is selected. Further, the partial load information of the estimation target consumer is estimated by using the selected representative model selected. Thereby, the partial load information of the estimation target consumer is accurately estimated.

Objects, features, aspects, and advantages of the present invention will become more apparent by the following detailed description and the accompanying drawings.

FIG. 3 is a block diagram illustrating the load estimation device of the first embodiment. FIG. 5 is a diagram showing a data structure of load information and a feature amount handled by the load estimation device of the first embodiment. 5 is an explanatory diagram illustrating a process performed in the load estimation device according to the first embodiment. FIG. 5 is an explanatory diagram illustrating a process performed in the load estimation device according to the first embodiment. FIG. FIG. 3 is a block diagram illustrating a load ratio classification unit included in the load estimation device of the first embodiment. FIG. 3 is a diagram showing an example of load information handled by the load estimation device of the first embodiment by means of graphs and tables. 5 is an explanatory diagram illustrating primary clustering performed in the load estimation device according to the first embodiment. FIG. FIG. 5 is a diagram showing a data structure of load information, a feature amount, and a primary cluster number handled by the load estimation device of the first embodiment. 5 is an explanatory diagram illustrating an example of deriving a primary cluster number by a derivation formula performed in the load estimation device of the first embodiment. FIG. FIG. 3 is a block diagram illustrating a representative model preparation unit included in the load estimation device according to the first embodiment. 5 is a graph showing an example of data X and Y handled in the load estimation device of the first embodiment. 6 is a diagram illustrating an example of a distance dist between each time series point of data X and each time series point of data Y, which is handled in the load estimation device of the first embodiment, by a table. FIG. FIG. 6 is a diagram illustrating an example of a cost cost required for associating each time series point of the data X and each time series point of the data Y, which is handled in the load estimation device of the first embodiment, with a table. In order to make the time series points of the data X corresponding to the time series points of the data Y, and the time series points of the data X correspond to the time series points of the data Y, which are handled in the load estimation apparatus of the first embodiment, respectively. It is a figure which illustrates the example of the required cost by a table. 5 is an explanatory diagram illustrating secondary clustering performed in the load estimation device according to the first embodiment. FIG. FIG. 5 is a diagram illustrating an example of a degree of similarity between total load information included in load information, which is handled by the load estimation device according to the first embodiment. 5 is an explanatory diagram illustrating third-order clustering performed in the load estimation device of the first embodiment. FIG. The time series point of the total load included in the first load information handled by the load estimation device of the first embodiment is the time series point of the data Y shown in FIG. 14, and the total time point included in the second load information. When the time series point of the load is the time series point of the data X shown in FIG. 14, the time series point of the device-specific load of the device A included in the first load information and the device included in the second load information It is a figure which illustrates the example of the time series point of the load according to the apparatus of A by a table. FIG. 3 is a diagram illustrating an example of a representative model handled by the load estimation device according to the first embodiment with a bar graph and a table. FIG. 3 is a diagram illustrating an example of a representative model handled by the load estimation device according to the first embodiment with a bar graph and a table. 3 is a block diagram illustrating a representative model acquisition unit included in the load estimation device according to the first embodiment. FIG. FIG. 5 is a diagram illustrating an example of a reference model, a non-reference model, and correspondence information handled by the load estimation device according to the first embodiment with a table. FIG. 3 is a block diagram illustrating an estimation unit included in the load estimation device according to the first embodiment. It is a figure which illustrates the example of the total load information of the estimation object house handled by the load estimation apparatus of Embodiment 1 by a table. The figure which illustrates the example of the information extracted from the total load information and the feature-value of the estimation object house handled by the load estimation apparatus of Embodiment 1, and the primary by the derivation|leading-out formula in the load estimation apparatus of Embodiment 1. It is a figure including an explanatory view explaining an example of derivation of a cluster number. FIG. 5 is a diagram illustrating an example of total load information of an estimation target handled by the load estimation device of the first embodiment with a line graph. FIG. 6 is a diagram illustrating an example of a selected representative model handled by the load estimation device according to the first embodiment by a stacked bar graph. It is a figure which overlaps and is illustrated so that the example of the total load information of an estimation object handled by the load estimation apparatus of Embodiment 1 and a selection representative model may be compared. FIG. 5 is a diagram illustrating an example of correspondence relationship information used when the correspondence relationship information for breakdown calculation used in the load estimation device according to the first embodiment is not used, using a table. FIG. 5 is a diagram illustrating, by a stacked bar graph, load information for each device estimated when the correspondence information for breakdown calculation handled in the load estimation device of the first embodiment is not used. FIG. 5 is a diagram illustrating an example of correspondence information for breakdown calculation handled in the load estimation device according to the first embodiment by a table. FIG. 6 is a diagram illustrating load information for each device estimated when using the correspondence information for breakdown calculation used in the load estimation device of the first embodiment. 7 is a block diagram illustrating a load estimation device according to a second embodiment. FIG. FIG. 9 is a block diagram illustrating a load ratio classification unit provided in the load estimation device of the second embodiment. FIG. 16 is an explanatory diagram illustrating an example of deriving a primary cluster number by a derivation formula performed in the load estimation device of the second embodiment. FIG. 7 is a block diagram illustrating an estimation unit included in the load estimation device according to the second embodiment. The figure which shows the example of the information extracted from the total load information and the feature-value of the estimation object house handled by the load estimation apparatus of Embodiment 2, and the temperature information by table, and derivation|leading-out in the load estimation apparatus of Embodiment 2. It is a figure containing an explanatory view explaining an example of derivation of a primary cluster number by a formula. FIG. 9 is a block diagram illustrating a load estimation device according to a third embodiment. FIG. 14 is a block diagram illustrating a load ratio classification unit provided in the load estimation device of the third embodiment. FIG. 16 is an explanatory diagram illustrating an example of deriving a primary cluster number by a derivation formula performed in the load estimation device of the third embodiment. FIG. 16 is a block diagram illustrating an estimation unit included in the load estimation device of the third embodiment. The figure which shows the example of the information extracted from the total load information and the feature-value of the estimation object housing handled by the load estimation apparatus of Embodiment 3, and the weather information by a table, and derivation|leading-out in the load estimation apparatus of Embodiment 3. It is a figure containing an explanatory view explaining an example of derivation of a primary cluster number by a formula.

Embodiment 1
1.1 Load Estimating Device FIG. 1 is a block diagram illustrating the load estimating device according to the first embodiment.

The load estimation apparatus 1000 according to the first embodiment illustrated in FIG. 1 includes a load ratio classification unit 1020, a representative model preparation unit 1021 and an estimation unit 1022. The load estimation device 1000 may include components other than these components.

The load estimation device 1000 is a computer that executes the installed program. The load ratio classification unit 1020, the representative model preparation unit 1021, and the estimation unit 1022 have functions realized by the computer executing the program. All or some of the functions may be realized by hardware that does not execute the program.

The load estimation apparatus 1000 acquires N pieces of load information 1040 and N pieces of characteristic amount 1041 from the monitoring house 1060, and acquires total load information 1080 and the amount of characteristic 1081 from the estimation target house 1100. N is an integer of 2 or more. Each of the N pieces of load information 1040 is load information for one day. The total load information 1080 is total load information for one day. The N pieces of load information 1040 and the N pieces of characteristic amount 1041 may be acquired from a consumer other than the house. The total load information 1080 and the characteristic amount 1081 may be acquired from a consumer other than the house. The N pieces of load information 1040 may or may not be acquired from N pieces of monitoring houses 1060 which are different from each other. When the N pieces of load information 1040 are not obtained from the N pieces of monitoring houses 1060 which are different from each other, the N pieces of load information 1040 include load information of a plurality of days obtained from one monitoring house. For example, the N pieces of load information 1040 may be N=A*B pieces of load information acquired by acquiring the load information for B days from each of the A monitoring homes 1060. In this case, the N feature quantities 1041 are N=A*B feature quantities acquired by acquiring B feature quantities from each of the A monitoring homes 1060.

In addition, the load estimation apparatus 1000 determines the N load information 1040 and the N feature amounts 1041 of the acquired monitoring house 1060, the total load information 1080 and the feature amount 1081 of the estimation target house 1100 for each device of the estimation target house 1100. The load information 1120 is estimated.

Further, the load estimation device 1000 outputs the estimated device-specific load information 1120 of the estimation target house 1100 to the service providing system 1140 and the demand prediction device 1141. The load information for each device 1120 may be output to an external device other than the service providing system 1140 and the demand prediction device 1141. The load information for each device 1120 may be used inside the load estimation apparatus 1000 without being output to an external device.

The service providing system 1140 provides various services to the estimation target house 1100 based on the input device-based load information 1120 of the estimation target house 1100.

The demand prediction apparatus 1141 predicts the demand for electric energy of the estimation target house 1100 based on the input device-specific load information 1120 of the estimation target house 1100.

“Load” is a general term for consumed electric energy, and is typically power consumption.

A “customer” is a unit of community that shares the consumption of electrical energy, typically a house. However, in the following, for convenience, the “consumer” may be treated as a person who needs to be supplied with electric energy, receives electric energy, and consumes the supplied electric energy.

1.2 Information Handled by Load Estimating Apparatus FIG. 2 is a diagram showing a data structure of load information and a feature amount handled by the load estimating apparatus of the first embodiment.

Each piece of load information 1160, which is each of the N pieces of load information 1040 of the monitoring house 1060, includes total load information 1180 and device-specific load information 1181, as shown in FIG. The N feature quantities 1041 of the monitoring house 1060 include the feature quantity 1200 associated with each load information 1160 illustrated in FIG. 2. Therefore, the N feature quantities 1041 are associated with the N load information 1040, respectively.

The total load information 1180 shows a time change in one day of the total load indicating the total amount of electric energy consumed by the monitoring house where the load information 1160 is acquired. Total load is shown hourly.

One day may be replaced by a shorter or longer period than one day. One hour may be replaced by a time interval shorter or longer than one hour. For example, 1 hour may be replaced with 1 minute, 30 minutes, or the like.

The total load information 1180 may include information indicating a temporal change in the amount of power generation indicating the magnitude of electric energy generated by solar power generation.

The device-specific load information 1181 indicates a temporal change in a day of the plurality of device-specific loads indicating the magnitude of electric energy consumed by each of the plurality of devices in the monitoring house from which the load information 1160 is acquired. A plurality of device-specific loads are shown every hour.

One day may be replaced by a shorter or longer period than one day. One hour may be replaced by a time interval shorter or longer than one hour. For example, 1 hour may be replaced with 1 minute, 30 minutes, or the like.

The device-specific load information 1181 may be replaced with partial load information other than the device-specific load information 1181. For example, the device-specific load information 1181 may be replaced with the application-specific load information. The application-specific load information indicates a temporal change of a plurality of application-specific loads indicating the magnitude of electric energy consumed by the plurality of applications of the monitoring house from which the load information 1160 is acquired. The equipment includes a kitchen, a living outlet, an induction heating (IH) cooking heater, a washing machine, a dishwasher, and the like. Applications include hot water supply, kitchen, entertainment, and air conditioning.

The load by device included in the load information by device 1181 is measured by the measuring device installed in the monitoring house where the load information 1160 is acquired. The measuring device may be inserted into each branch circuit of the distribution board, or may be incorporated in the outlet tap. The load for each device may be measured by the load measuring function of the device. The load by device measured by the home energy management system (HEMS) may be acquired. The measuring device may be inserted into the main trunk circuit of the distribution board, and the device-specific load estimated from the measurement result of the inserted measuring device may be acquired. However, the estimated load by device must have high accuracy.

The device-based load information 1181 may include information indicating a change over time in the amount of power generation indicating the magnitude of electric energy generated by solar power generation.

The feature amount 1200 indicates the feature of the monitoring house where the load information 1160 is acquired. The characteristics of the monitoring house include the characteristics of the monitoring house itself and the characteristics of the resident living in the monitoring house. The characteristics of the monitoring house itself are addresses, latitude and longitude, age, floor area, etc., but are not limited to these. The characteristics of the residents living in the monitoring house are the number of residents, family structure, age, etc., but are not limited to these.

The total load information 1080 of the estimation target house 1100 indicates a temporal change in one day of the total load indicating the total amount of electric energy consumed by the estimation target house 1100. Total load is shown hourly.

One day may be replaced by a shorter or longer period than one day. For example, one day may be replaced by one week. One hour may be replaced by a time interval shorter or longer than one hour. For example, 1 hour may be replaced with 1 minute, 30 minutes, or the like.

The total load included in the total load information 1080 of the estimation target house 1100 is obtained from the measurement amount measured by the smart meter installed in the estimation target house 1100. However, the amount measured by the smart meter includes the amount of electric power generated by solar power generation, which indicates the amount of electric energy generated, and the amount of electric energy input and output by the storage battery. Has been added. For this reason, when the total load is obtained from the measured amount measured by the smart meter, the power generation amount, the input/output amount, and the like measured by other than the smart meter are subtracted from the measured amount measured by the smart meter. The estimated amount of power generation may be subtracted instead of the measured amount of power generation. The total load to be estimated may be obtained from other than the measured amount measured by the smart meter.

The feature amount 1081 of the estimation target house 1100 indicates the feature of the estimation target house 1100. The features of the estimation target house 1100 include the features of the estimation target house 1100 itself, the features of a resident who lives in the estimation target house 1100, and the like. The features of the estimation target house 1100 itself are an address, latitude and longitude, a building age, a floor area, etc., but are not limited thereto. The characteristics of the resident living in the estimation target housing 1100 are the number of residents, family structure, age, etc., but are not limited to these.

The feature amount 1081 of the estimation target house 1100 is acquired from the device to which the feature amount 1081 is input when the device-specific load information 1120 of the estimation target house 1100 is estimated. For example, the feature amount 1081 is input to a device such as a personal computer in the estimation target house 1100 when the device-specific load information 1120 is estimated, and is acquired from the device via the Internet. The characteristic amount 1081 may be stored in advance in a device that stores the characteristic amount 1081, and may be acquired from the device when the device-specific load information 1120 is estimated. For example, the feature amount 1081 is input in advance to a device such as a personal computer in the estimation target house 1100, transferred from the device via the Internet and stored in a device such as a server, and the device-specific load information 1120 is estimated. At that time, it is acquired from the device.

The load information for each device is not acquired from the estimation target house 1100. Therefore, it is not necessary to install a measuring device for measuring the load for each device in the estimation target house 1100.

The device-based load information 1120 of the estimation target house 1100 indicates a temporal change of a plurality of device-based loads indicating the magnitude of electric energy consumed by the plurality of devices of the estimation target house 1100. When the device-specific load information 1181 is replaced with partial load information other than the device-specific load information 1181, the device-specific load information 1120 is also replaced with partial load information other than the device-specific load information 1120.

The magnitude of electric energy is expressed by the amount of electric power. The magnitude of electric energy may be expressed by something other than electric power. For example, the magnitude of electric energy may be represented by electric power, electric current, or the like. The load for each device may be expressed by the ratio of the total load.

1.3 Processes Performed by Load Estimating Device FIGS. 3 and 4 are explanatory diagrams illustrating processes performed by the load estimating device according to the first embodiment.

The load ratio classification unit 1020 acquires N load information 1040 and N feature amounts 1041 illustrated in FIG. 3 from the monitoring house 1060.

Further, as illustrated in FIG. 3, the load ratio classification unit 1020 performs primary clustering on the N pieces of load information 1040 and the N pieces of characteristic amount 1041 of the acquired monitoring house 1060, and the N pieces of load are obtained. The information 1040 and the N feature quantities 1041 are classified into M_1 primary clusters 1220, and N(m_1) load information 1260 belonging to each primary cluster 1240 that is each of the M_1 primary clusters 1220 and N(m_1) feature quantities 1261 are specified. The specified N(m_1) feature quantities 1261 are associated with the specified N(m_1) load information 1260, respectively. M_1 is smaller than N and is an integer of 2 or more except in exceptional cases. The primary clustering is performed so that the N(m_1) device-specific load ratios calculated for the N(m_1) load information 1260 belonging to each primary cluster 1240 are similar to each other. Therefore, each primary cluster 1240 is a device-based load ratio cluster to which N(m_1) pieces of load information 1260 having similar device-based load ratios belong. The device-specific load ratio calculated for each load information 1160 indicates the ratio of a plurality of device-specific loads included in each load information 1160 to the total load included in each load information 1160.

Further, as shown in FIG. 3, the load ratio classification unit 1020 learns N(m_1) load information 1260 and N(m_1) feature amounts 1261 belonging to each primary cluster 1240, and gives them. A derivation formula 1280 for deriving a primary cluster from the total load information and the given feature amount is created.

As shown in FIGS. 3 and 4, the representative model preparation unit 1021 performs secondary clustering on the specified N(m_1) pieces of load information 1260 to specify the specified N(m_1) pieces of load. The information 1260 is classified into M_2 secondary clusters 1300, and N(m_2) load information 1340 belonging to each secondary cluster 1320 that is each of the M_2 secondary clusters 1300 is specified. As a result, N pieces of load information 1040 of the monitoring house 1060 are classified into M_1×M_2 secondary clusters, and N(m_2) pieces belonging to each secondary cluster 1320 that is each of the M_1×M_2 secondary clusters. Load information 1340 is specified. M_2 is smaller than N(m_1) and is an integer of 2 or more except in exceptional cases. The secondary clustering is performed such that the N(m_2) pieces of total load information included in the N(m_2) pieces of load information 1340 belonging to each secondary cluster 1320 are similar to each other.

As illustrated in FIG. 4, the representative model preparation unit 1021 performs third-order clustering on the specified N(m_2) pieces of load information 1340 to specify the specified N(m_2) pieces of load information. 1340 is classified into M_3 tertiary clusters 1360, and N(m_3) pieces of load information 1400 belonging to each tertiary cluster 1380, which is each of the M_3 tertiary clusters 1360, are specified. As a result, the N load information 1040 of the monitoring house 1060 is classified into M_1×M_2×M_3 tertiary clusters, and N belonging to each tertiary cluster 1380 that is each of the M_1×M_2×M_3 tertiary clusters. (m_3) pieces of load information 1400 are specified. M_3 is N(m_2) or less, and is an integer of 2 or more except for exceptional cases. The tertiary clustering is performed such that the N(m_3) pieces of load information for each device included in the N(m_3) pieces of load information 1400 belonging to each of the tertiary clusters 1380 are similar to each other.

In addition, the representative model preparation unit 1021 creates M_1* by representing each representative model 1420 representing N(m_3) pieces of load information 1400 belonging to each tertiary cluster 1380, as shown in FIG. Prepare M_2*M_3 representative models 1440.

The estimation unit 1022 acquires the total load information 1080 and the feature amount 1081 illustrated in FIG. 3 from the estimation target house 1100.

Further, as illustrated in FIG. 3, the estimation unit 1022 sets the selected primary cluster 1460 derived from the total load information 1080 and the feature amount 1081 of the estimation target house 1100 by the derivation formula 1280 to M_1 primary clusters 1220. Select from. Further, as illustrated in FIG. 4, the estimation unit 1022 selects the selected representative model 1520 from the two or more representative models 1500 that represent the load information 1480 belonging to the selected primary cluster 1460.

Furthermore, the estimation unit 1022 estimates the device-specific load information 1120 of the estimation target house 1100 from the total load information 1080 of the estimation target house 1100 using the selected selection representative model 1520.

1.4 Load Ratio Classifying Unit FIG. 5 is a block diagram illustrating a load ratio classifying unit included in the load estimating device according to the first embodiment.

As shown in FIG. 5, the load ratio classification unit 1020 includes a load information storage unit 1540, a device-based load ratio calculation unit 1541, a primary cluster number determination unit 1542, a learning data classification unit 1543, and a derivation formula creation unit 1544. Prepare

1.5 Load Information Storage Unit The load information storage unit 1540 acquires N pieces of load information 1040 and N pieces of characteristic amount 1041 from the monitoring house 1060, and acquires the obtained N pieces of load information 1040 and N pieces of characteristic amount 1041. Memorize

1.6 Example of Load Information FIG. 6 is a diagram showing an example of load information handled by the load estimating apparatus according to the first embodiment with graphs and tables.

The load information illustrated in FIG. 6(a) is acquired from “customer 1”, and the total load information indicated by a line graph and the entire components of the stacked bar graph and the multiple components in the stacked bar graph are obtained. The load information for each device is included.

The load information illustrated in FIG. 6B is acquired from “customer 1” and includes total load information and device-specific load information shown in the table.

1.7 Device-Specific Load Ratio Calculation Unit The device-specific load ratio calculation unit 1541, for each load information 1160, is a ratio of a plurality of device-specific loads included in the device-specific load information 1181 to the total load included in the total load information 1180. The load ratio by device is calculated. As a result, N device-specific load ratios 1560 are calculated for each of the N load information 1040. The calculated load ratio by device may be a load ratio by device at each time, or may be a load ratio by device for one day. The time segment when the device-specific load ratio is calculated is arbitrarily set to be equal to or greater than the time segment in which the device-specific load information 1181 indicates the device-specific load.

1.8 Primary Cluster Number Determining Section The primary cluster number determining section 1542 determines the N load information 1040 and the N load information 1040 of the monitoring house 1060 according to the size of the information indicating the calculated N load ratios 1560 by device. The number of primary clusters M_1, which is the number of primary clusters 1220 into which the feature amount 1041 of FIG. The number of primary clusters M_1 may be determined according to the input operation of the user of the load estimation device 1000. The number of primary clusters M_1 may be determined by analyzing the N pieces of load information 1040.

1.9 Learning Data Classifying Unit FIG. 7 is an explanatory diagram illustrating the primary clustering performed in the load estimating device according to the first embodiment. FIG. 8 is a diagram showing a data structure of load information, a feature amount, and a primary cluster number handled by the load estimation device according to the first embodiment.

The learning data classification unit 1543 includes a similarity calculation unit 1580 for primary clustering and a primary clustering unit 1581 as illustrated in FIG.

As shown in FIG. 7, the primary clustering similarity calculation unit 1580 calculates the primary clustering similarity 1600 among the calculated N device-based load ratios 1560.

As shown in FIG. 7, the primary clustering unit 1581 is stored in the load information storage unit 1540 according to the primary clustering similarity 1600 calculated by the primary clustering similarity calculating unit 1580. N load information 1040 and N feature amounts 1041 of the monitoring house 1060 are classified into M_1 primary clusters 1220, and each of the M_1 primary clusters 1220 belongs to each primary cluster 1240, N( The load information 1260 of m_1) and the feature amount 1261 of N(m_1) are specified. The N(m_1) feature amounts 1261 are associated with the N(m_1) load information 1260, respectively. The N(m_1) pieces of load information 1260 are at least one piece of load information. The N(m_1) feature quantities 1261 are at least one feature quantity. The primary clustering is performed so that the N(m_1) device-specific load ratios calculated for the N(m_1) load information 1260 belonging to each primary cluster 1240 are similar to each other. The time division when the N pieces of load information 1040 are classified into the M_1 primary clusters 1220 is the same as the time division when the device-specific load ratio is calculated.

Further, the primary clustering unit 1581 causes the load information storage unit 1720 to store N(m_1) pieces of load information 1260 belonging to each primary cluster 1240, as illustrated in FIG.

Further, as shown in FIG. 8, the primary clustering unit 1581 identifies the primary cluster number 1620 for identifying the primary cluster to which each load information 1160 and the feature amount 1200 belongs from other primary clusters, each load information 1160. Correspond to. As a result, as shown in FIG. 5, N pieces of classified load information 1640 are obtained.

1.10 Derivation Formula Generation Unit As shown in FIG. 7, the derivation formula generation unit 1544 stores N(m_1) load information 1260 and N(m_1) feature amounts 1261 belonging to each primary cluster 1240. After learning, a derivation formula 1280 for deriving the primary cluster number 1662 from the given total load information 1660 and the given feature amount 1661 is created. The derived primary cluster number 1662 identifies the primary cluster. Therefore, the derived derivation formula 1280 derives a primary cluster from the given total load information 1660 and the given feature amount 1661.

The derivation formula 1280 is created by learning N(m_1) pieces of load information 1260 and N(m_1) pieces of feature amount 1261 belonging to each primary cluster 1240. Therefore, the primary cluster number 1662 derived by the derivation formula 1280 is the primary cluster number that identifies the primary cluster to which the total load information and the feature amount most similar to the given total load information 1660 and the feature amount 1661 belong. Is. The primary cluster to which the total load information belongs is a primary cluster to which the load information including the total load information belongs. If the total load information 1660 is the total load information included in the load information belonging to the specific primary cluster, and the characteristic amount 1661 is the characteristic amount associated with the load information, the derivation formula is The primary cluster number 1662 derived by 1280 is the primary cluster number that identifies the particular primary cluster.

The derivation formula 1280 may be created using a method of multivariately analyzing the N(m_1) load information 1260 and the N(m_1) feature amount 1261, or using the load estimation apparatus 1000. It may be performed based on the total load information 1660 and the threshold value of the feature amount 1661 set according to the input operation by the person. Methods for multivariately analyzing N(m_1) load information 1260 and N(m_1) feature quantities 1261 include a method using a decision tree and a method using a support vector machine.

Further, the derivation formula creating unit 1544 stores the created derivation formula 1280 in the derivation formula storage unit 1680.

1.11 Derivation Example of Primary Cluster Number FIG. 9 is an explanatory diagram illustrating an example of derivation of a primary cluster number by a derivation formula performed in the load estimation device of the first embodiment.

In the derivation of the primary cluster number by the derivation formula 1280, the given total load information 1660 and the given feature amount 1661 are analyzed, and the information that influences the derived primary cluster number 1662 is extracted. As shown in FIG. 5, the nodes 1700, 1701, 1702, and 1703 determine the magnitude of the extracted information and the reference value, whether the extracted information meets the conditions, and the like. Then, based on the result, the derived primary cluster number 1662 is selected from the primary cluster number candidates 1710, 1711, 1712, 1713 and 1714. The extracted information may be information directly extracted from the total load information 1660 and the feature amount 1661, or may be information extracted by processing the total load information 1660 and the feature amount 1661. Good. The information extracted by processing the total load information 1660 is included in the total load information 1660 including the total load statistical values such as the maximum value, the minimum value, and the average value of the total load included in the total load information 1660 for one hour. It is a numerical value or the like obtained by performing a Fourier transform on the total load.

1.12. Representative Model Preparation Unit FIG. 10 is a block diagram illustrating a representative model preparation unit included in the load estimation device according to the first embodiment.

As shown in FIG. 10, the representative model preparation unit 1021 includes a load information storage unit 1720, a correspondence calculation unit 1721 for clustering, a similarity calculation unit 1722 for secondary clustering, a secondary clustering unit 1723, and a tertiary clustering. A similarity calculation unit 1724 for clustering, a tertiary clustering unit 1725, a representative model acquisition unit 1726, and a representative model storage unit 1727 are provided.

1.13 Load Information Storage Unit The load information storage unit 1720 stores N(m_1) pieces of load information 1260 belonging to each primary cluster 1240.

1.14 Correspondence Calculation Unit for Clustering The correspondence relation calculation unit 1721 for clustering stores all load information in N (m_1) load information 1260 belonging to each primary cluster 1240 stored in the load information storage unit 1720. For each load information pair that is each of the pairs, the total load information included in the first load information forming each load information pair and the total load information included in the second load information forming each load information pair Correspondence relationship information indicating a time series correspondence relationship between the two is calculated. As a result, the correspondence relationship information 1740 for clustering indicating the time-series correspondence relationship between the N (m_1) pieces of total load information included in the N (m_1) pieces of load information 1260 belonging to each primary cluster 1240. Is calculated.

The correspondence relationship calculation unit 1721 calculates the correspondence relationship information indicating the time-series correspondence relationship between the total load information included in the first load information and the total load information included in the second load information. , The total load included in the first load information so that the plurality of time series points of the total load included in the first load information approximate the plurality of time series points of the total load included in the second load information. A plurality of time series points of the total load included in the second load information corresponding to the plurality of time series points of the load are specified.

1.15 Example of Correspondence Calculation Method Data used for calculation of time-series correspondence between total load information included in the first load information and total load information included in the second load information An example of the calculation method of the time series correspondence between X and the data Y will be described below.

The time-series correspondence between the data X and the data Y is obtained while the data X and the data Y are matched with each other. The matching between the data X and the data Y is performed by, for example, the dynamic time warping method (DTW).

Each of the data X and the data Y indicates the change with time of the value. In the following, the time series points x_t(1),x_t where the data X indicates the values at the times t(1), t(2), t(3), t(4), t(5) and t(6), respectively. (2), x_t(3), x_t(4), x_t(5) and x_t(6), and the data Y has time t(1), t(2), t(3), t(4). , t(5) and t(6) have time series points y_t(1), y_t(2), y_t(3), y_t(4), y_t(5) and y_t(6), respectively. To do.

FIG. 11 is a graph showing an example of data X and Y handled by the load estimation device according to the first embodiment.

11(a) and 11(b) show examples of data X and Y, respectively. The data X and Y shown in FIGS. 11A and 11B are simplified so that the value indicated by each time series point of the data X and Y becomes 0 or 1.

When the data X and the data Y are matched, the time series points x_t(1), x_t(2), x_t(3), x_t(4), x_t(5) and x_t(6) of the data X are compared. The distance dist between each and the time series points y_t(1), y_t(2), y_t(3), y_t(4), y_t(5) and y_t(6) of the data Y is calculated.

The distance dist(x_t(i),y_t(j)) between the time series point x_t(i) of the data X and the time series point y_t(j) of the data Y is calculated by the equation (1).

FIG. 12 is a diagram illustrating an example of a distance dist between each time series point of the data X and each time series point of the data Y, which is handled in the load estimating apparatus according to the first embodiment.

Then, using the calculated distance dist, each of the time series points x_t(1), x_t(2), x_t(3), x_t(4), x_t(5) and x_t(6) of the data X The cost cost required to associate each of the time series points y_t(1), y_t(2), y_t(3), y_t(4), y_t(5) and y_t(6) of the data Y is calculated. It The cost cost(x_t(i),y_t(j)) required to associate the time-series point x_t(i) of the data X and the time-series point y_t(j) of the data Y is given by the equations (2), ( It is calculated by 3), (4) and (5).

Formula (2) calculates the cost required to associate the time series point x_t(1) at time t(1) of the data X and the time series point y_t(1) at time t(1) of the data Y. Is an arithmetic expression for The cost is an initial value. Formula (3) is the time of the data Y at each of the time series points x_t(2), x_t(3), x_t(4), x_t(5) and x_t(6) other than the time t(1) of the data X. It is an arithmetic expression for obtaining the cost required to correspond to the time series point y_t(1) at t(1). Formula (4) is the time series point x_t(1) at time t(1) of data X and the time series points y_t(2), y_t(3), y_t(4) at times other than time t(1) of data Y. , y_t(5) and y_t(6). Formula (5) is the time of the data Y at each time series point x_t(2), x_t(3), x_t(4), x_t(5) and x_t(6) other than the time t(1) of the data X. An arithmetic expression for finding the cost required to correspond to each of the time series points y_t(2), y_t(3), y_t(4), y_t(5) and y_t(6) other than t(1) Is. Factors min(cost(x_t(i),y_t(j-1)),cost(x_t(i-1),y_t(j)),cost(x_t(i-1),y_t included in equation (5) (j-1))) is cost(x_t(i),y_t(j-1)), cost(x_t(i-1),y_t(j)) and cost(x_t(i-1),y_t( j-1)) is a function that returns the minimum argument of the three arguments.

FIG. 13 is a diagram illustrating, by a table, an example of a cost cost required to associate each time series point of the data X with each time series point of the data Y, which is handled by the load estimation device according to the first embodiment. is there.

Then, the time corresponding to each time series point of time series point y_t(1), y_t(2), y_t(3), y_t(4), y_t(5) and y_t(6) of data Y. The series points are extracted from the time series points x_t(1), x_t(2), x_t(3), x_t(4), x_t(5) and x_t(6) of the data X. Extraction is performed at the time series points x_t(1), x_t(2), x_t(3), x_t(4), x_t(5) and x_t(6) of the data X and the time series points at the end of the data Y. Costs required to match y_t(6) cost(x_t(1),y_t(6)),cost(x_t(2),y_t(6)),cost(x_t(3),y_t(6)) ,cost(x_t(4),y_t(6)), cost(x_t(5),y_t(6)) and cost(x_t(6),y_t(6)), the minimum cost min(cost(x_t (1),y_t(6)),cost(x_t(2),y_t(6)),cost(x_t(3),y_t(6)),cost(x_t(4),y_t(6)),cost (x_t(5),y_t(6)),cost(x_t(6),y_t(6))) is calculated by using the distance required in the process of calculation. The extraction is performed so that the cost required for associating the extracted time series points of the data X with the respective time series points of the data Y is minimized.

As a result, the six time series points y_t(1), y_t(2), y_t(3), y_t(4), y_t(5), and y_t(6) of the data Y have six data points. Is specified, and the correspondence relationship between the data X and the data Y in time series is specified. Correspondence relationship information indicating a time-series correspondence relationship between the data X and the data Y is six time series points y_t(1), y_t(2), y_t(3), y_t(4) of the data Y. , y_t(5) and y_t(6), respectively, including information indicating six time series points of the data X.

6 time series of data Y 6 time series of data X respectively corresponding to y_t(1), y_t(2), y_t(3), y_t(4), y_t(5) and y_t(6) The time change of the value at the point is the time of the value at the six time series points y_t(1), y_t(2), y_t(3), y_t(4), y_t(5) and y_t(6) of the data Y. Approximate to change.

Instead of the time series point y_t(6) serving as the end point, time series points y_t(2), y_t(3), y_t(4) or y_t(5) not serving as the end point may be used. The distance dist(x_t(1),y_t(1)) does not have to be included in the distance required for calculating the minimum cost. When there are a plurality of costs that can be used in the process of calculating the minimum cost, a cost arbitrarily selected from the plurality of costs may be used. For example, in the example of the cost cost shown in FIG. 13, a cost arbitrarily selected from the costs cost(x_t(5),y_t(5)) and cost(x_t(6),y_t(5)) is used. May be.

In matching, the time series point y_t(1) that is the start point of the data Y does not have to be associated with the time series point x_t(1) that is the start point of the data X, and the time series point y_t that is the end point of the data Y is not required. The time series point x_t(6), which is the end point of the data X, may not be associated with (6).

FIG. 14 shows a time series point of data X, which corresponds to a time series point of data Y, and a time series point of data X and a time series point of data Y, which are handled in the load estimation apparatus of the first embodiment. It is a figure which illustrates the example of the cost required to correspond by a table.

When the above calculation method is used to calculate the time-series correspondence between the total load information included in the first load information and the total load information included in the second load information, the first load The total load information included in the information and the total load information included in the second load information are converted into data Y and X, respectively, and are included in the total load information (data Y) included in the first load information and the second load information. A time-series correspondence relationship with the included total load information (data X) is calculated.

The time series points of the total load included in the first load information (time series points of the data Y) and the time series points of the total load included in the second load information (time series points of the data X) are associated with each other. The cost required for this has a characteristic that it becomes smaller as the degree of similarity between the total load information included in the first load information and the total load information included in the second load information increases. Therefore, the cost can be used as the degree of similarity between the total load information included in the first load information and the total load information included in the second load information.

When the above calculation method is used to calculate the time-series correspondence between the total load information included in the first load information and the total load information included in the second load information, the first load Normalization may be performed such that the total load range included in the information and the total load range included in the second load information are a common range, for example, a range from 0 to 1. The data X is divided into a plurality of time zone data belonging to a plurality of time zones, the data Y is divided into a plurality of time zone data belonging to a plurality of time zones, and the data X and Y belonging to each time zone are divided. Correspondence may be calculated between the two pieces of time zone-based data that are respectively configured. At that time, the numerical value subjected to the processing using the weighting according to the time zone may be used for determining the correspondence. Whether or not to perform the process and the weight necessary for performing the process may be manually input to the load estimation device 1000 by the user, or automatically determined by a computer such as a personal computer (PC). May be. The user can determine whether or not to perform the process and the weight necessary for performing the process according to the purpose of use of the load estimation apparatus 1000. The PC can determine whether or not to perform the process based on the purpose of use selected by the user and the weight necessary for performing the process.

The matching may be performed by a method other than the DTW method.

1.16 Similarity Calculation Unit for Secondary Clustering FIG. 15 is an explanatory diagram illustrating secondary clustering performed in the load estimation device according to the first embodiment.

The similarity calculation unit 1722 for secondary clustering uses the load information that is each of all load information pairs in the N (m_1) load information 1260 belonging to each primary cluster 1240 stored in the load information storage unit 1720. For each pair, the similarity between the total load information included in the first load information forming each load information pair and the total load information included in the second load information forming each load information pair is calculated. Thereby, as shown in FIG. 15, the similarity 1780 for secondary clustering among the N(m_1) total load information 1760 included in each of the N(m_1) load information 1260 is calculated. ..

The similarity calculation unit 1722 calculates the similarity between the total load information included in the first load information and the total load information included in the second load information, and the correspondence calculation unit 1721 for clustering. The correspondence information 1740 for clustering calculated by is referred to.

The similarity calculation unit 1722 refers to the correspondence information 1740 to refer to the plurality of total loads included in the second load information, which correspond to the plurality of time series points of the total load included in the first load information, respectively. Specify the time series points of.

The similarity calculation unit 1722 calculates the similarity between the plurality of time series points of the total load included in the first load information and the plurality of time series of the total load included in the second load information. , The similarity between the total load information included in the first load information and the total load information included in the second load information.

The calculation method described in “1.15 Example of Calculation Method of Correspondence Relationship” is used, and a plurality of time series points of the total load included in the first load information and a plurality of total loads included in the second load information are used. When the cost required to respectively correspond to the time series of is used as the similarity between the total load information included in the first load information and the total load information included in the second load information, The similarity calculation unit 1722 is omitted, and the correspondence calculation unit 1721 also serves as the similarity calculation unit for primary clustering.

1.17 Secondary Clustering Unit As shown in FIG. 15, the secondary clustering unit 1723 follows the load information according to the secondary clustering similarity 1780 calculated by the secondary clustering similarity calculating unit 1722. The N(m_1) pieces of load information 1260 belonging to each primary cluster 1240 stored in the storage unit 1720 are classified into M_2 secondary clusters 1300, and each secondary cluster that is each of the M_2 secondary clusters 1300. The N(m_2) pieces of load information 1340 belonging to 1320 are specified. The N(m_2) pieces of load information 1340 are at least one piece of load information. The secondary clustering is performed such that the N(m_2) total load information 1800 included in each of the N(m_2) load information 1340 belonging to each secondary cluster 1320 is similar to each other.

N(m-1) pieces of load information 1260 belonging to each of the first at least one primary clusters included in the M_1 primary clusters 1220 are one piece of load information. When the N(m_1) pieces of load information 1260 belonging to each of the included second at least one primary cluster are a plurality of load information, the similarity calculation unit 1724 determines that the second at least one primary cluster. The similarity 1780 for secondary clustering, which is the similarity between the N(m_1) pieces of total load information 1760 respectively included in the N(m_1) pieces of load information 1260 belonging to each of the above, is calculated.

FIG. 16 is a diagram illustrating an example of the degree of similarity between the total load information included in the load information, which is handled by the load estimation device according to the first embodiment.

In the example shown in FIG. 16, the total load included in the load information 1821 in the load information 1821, 1822, 1823, 1824, 1825 and 1826 to which the data numbers 1, 2, 3, 4, 5 and 6 are respectively given. The similarity between the information and the total load information included in the load information 1822 is high, and the similarity between the total load information included in the load information 1824 and the total load information included in the load information 1825 is high. In addition, the similarity between the total load information included in the load information of the cluster formed when the load information 1821 and 1822 are integrated and the total load information included in the load information 1823 is medium, and the load information The similarity between the total load information included in the load information of the cluster formed when 1824 and 1825 are integrated and the total load information included in the load information 1826 is medium. In addition, the total load information included in the load information of the cluster formed when the load information 1821, 1822, and 1823 are integrated and the load information of the cluster formed when the load information 1824, 1825, and 1826 are integrated are included. The degree of similarity with the total load information included is low.

In the secondary clustering, each load information pair which is each of all the load information pairs in the load information 1821, 1822, 1823, 1824, 1825 and 1826 is included in the first load information forming each load information pair. The similarity between the total load information and the total load information included in the second load information forming each load information pair is calculated, and the first load forming the load information pair having the highest similarity is calculated. The information and the second load information are made to belong to the same primary cluster.

The calculation method described in "1.15 Example of Correspondence Calculation Method" corresponds in time series between the total load information included in the first load information and the total load information included in the second load information. The cost required for calculating the relationship and corresponding to the plurality of time series points of the total load included in the first load information and the plurality of time series of the total load included in the second load information are When used as the degree of similarity between the total load information included in the first load information and the total load information included in the second load information, it is included in the first load information when the cost is small. It is determined that the similarity between the total load information and the total load information included in the second load information is high. In addition, the first load information and the second load information are integrated in the ascending order of cost.

In the simplest example, each load information pair that is each of all the load information pairs in the load information 1821, 1822, 1823, 1824, 1825, and 1826 is included in the first load information that configures each load information pair. The cost required to associate the plurality of time series points of the total load with the plurality of time series points of the total load included in the second load information forming each load information pair, and the calculated cost It is determined that the similarity between the two pieces of total load information respectively included in the first load information and the second load information forming the load information pair having the smallest?

The calculation of the similarity between the total load information included in the load information of the cluster formed when a plurality of load information is integrated and the total load information included in other load information is performed by the nearest neighbor method and the farthest method. Neighbor method, group average method, median method, center of gravity method, Ward method, etc.

The method of determining the load information of the cluster formed when a plurality of load information are integrated is arbitrary. For example, one piece of load information representing at least one piece of load information belonging to the formed cluster is set as the load information of the formed cluster.

In the secondary clustering, the number of secondary clusters into which the load information 1821, 1822, 1823, 1824, 1825 and 1826 are classified may be designated, or may be included in at least one load information belonging to the same secondary cluster. The degree of similarity between at least one total load information to be specified may be designated.

For example, when designation is made to increase the similarity between at least one total load information included in at least one load information belonging to the same primary cluster, the load information 1821, 1822, 1823, 1824, 1825 is specified. And 1826 are set to four primary clusters, and after the primary clustering is performed, the load information 1821 and 1822 belong to the first primary cluster, and the load information 1823 is the second primary cluster. It belongs to the next cluster, the load information 1824 and 1825 belong to the third primary cluster, and the load information 1826 belongs to the fourth primary cluster.

When the number of primary clusters into which the load information 1821, 1822, 1823, 1824, 1825, and 1826 is classified is set to two, at least one load information belonging to the same primary cluster is respectively included. After the degree of similarity between at least one total load information included is set to medium and the primary clustering is performed, the load information 1821, 1822 and 1823 belong to the first primary cluster and the load information 1824, 1825 and 1826 belong to the second primary cluster.

1.18 Similarity Calculation Unit for Third-Order Clustering FIG. 17 is an explanatory diagram illustrating third-order clustering performed in the load estimation device according to the first embodiment.

The degree-of-similarity calculation unit 1724 for tertiary clustering is each of all load information pairs in N(m_2) load information 1340 belonging to each secondary cluster 1320 that is each M_2 secondary cluster 1300. Regarding the load information pair, the similarity between the load information by device included in the first load information forming each load information pair and the load information by device included in the second load information forming each load information pair To calculate. As a result, the degree of similarity 1860 for tertiary clustering is calculated among the N(m_2) pieces of load information for each device 1840 included in the N(m_2) pieces of load information 1340 belonging to each secondary cluster 1320.

The N(m-2) load information 1340 belonging to each of the first at least one secondary clusters included in the M_2 secondary clusters 1300 is one load information, and the M_2 secondary clusters 1300 include When the N(m_2) pieces of load information 1340 belonging to each of the included second at least one secondary cluster are a plurality of load information, the similarity calculation unit 1724 determines that the second at least one primary cluster. The similarity degree 1860 for the third clustering, which is the similarity degree between the N(m_2) pieces of load information 1840 for each device included in the N(m_2) pieces of load information 1340 belonging to each of the above, is calculated.

The similarity calculation unit 1724 calculates the correspondence relationship for clustering when calculating the similarity between the device-specific load information included in the first load information and the device-specific load information included in the second load information. The correspondence information 1740 for clustering calculated by the unit 1721 is referred to.

By referring to the correspondence information 1740, the similarity calculation unit 1724 refers to the plurality of total loads included in the second load information, which correspond to the plurality of time series points of the total load included in the first load information, respectively. Specify the time series of.

In addition, the similarity calculation unit 1724 determines the plurality of time series points of the total load included in the first load information and the plurality of time series points of the device-specific load acquired at the same time and the second load information. Of the total load included in the first load information and the plurality of time series points of the device load acquired at the same time, the device load information included in the first load information and the second load The similarity is to the load information for each device included in the information.

FIG. 18 is a time series point of the total load included in the first load information handled by the load estimation apparatus of the first embodiment, which is the time series point of the data Y shown in FIG. When the time series point of the total load included in is the time series point of the data X shown in FIG. 14, the time series point of the device-specific load of the device A included in the first load information and the second load information. 5 is a diagram illustrating an example of a time series point of a load by device of device A included in FIG.

In the example illustrated in FIG. 18, the time series points yA_t(1), yA_t(2), yA_t(3), yA_t(4), yA_t(5 of the device-specific load of the device A included in the first load information. ) And yA_t(6), the time series points xA_t(1), xA_t(2), xA_t(2), xA_t(2), xA_t(5) of the load by device of the device A included in the second load information. And xA_t(6) are associated with each other. Time series points yA_t(1), yA_t(2), yA_t(3), yA_t(4), yA_t(5) and yA_t(6) are time series points y_t( of the total load included in the first load information. 1), y_t(2), y_t(3), y_t(4), y_t(5) and y_t(6) are acquired at the same time. The time series points xA_t(1), xA_t(2), xA_t(2), xA_t(2), xA_t(5) and xA_t(6) are the time series points x_t( of the total load included in the second load information. 1), x_t(2), x_t(2), x_t(2), x_t(5) and x_t(6) are acquired at the same time.

Error ε_A(xA_t(xA_t(i) between the time series point yA_t(j) of the device-specific load included in the first load information and the device-specific load xA_t(i) of the device A included in the second load information. i) and yA_t(j)) are calculated by the equation (6).

Device-specific load error ε_A between a plurality of time-series points of the device-specific load included in the first load information and a plurality of time-series points of the device-specific load included in the second load information. (X, Y) is a time series point of each of the plurality of time series points of the load of the device A included in the first load information, and each time series point is yA_t(j). The error ε_A(xA_t(i),yA_t(j)) is calculated with the time series point of the device-specific load of the device A corresponding to each time series point included in the load information of xA_t(i) calculated as It is calculated by summing a plurality of errors ε_A(xA_t(i), yA_t(j)) calculated for a plurality of time series points of the device-specific load of the device A included in the load information.

The device-specific load error ε_(X,Y) between the device-specific load information included in the first load information and the device-specific load information included in the second load information is defined as device A for each of the plurality of devices. It is calculated by calculating the device-specific load error ε_A(X,Y) and summing the plurality of device-specific load errors ε_A(X,Y) respectively calculated for the plurality of devices.

The calculated device-specific load error ε_A(X,Y) is smaller as the similarity between the device-specific load information included in the first load information and the device-specific load information included in the second load information is higher. Have the following characteristics. Therefore, the device-specific load error ε_A(X,Y) can be used as the degree of similarity between the device-specific load information included in the first load information and the device-specific load information included in the second load information.

1.19 Third-Order Clustering Unit As shown in FIG. 17, the third-order clustering unit 1725 uses M_2 pieces according to the third-order clustering similarity degree 1860 calculated by the third-order clustering similarity degree calculation unit 1724. N(m_2) pieces of load information 1340 belonging to each secondary cluster 1320, which is each of the secondary clusters 1300 of 3), is classified into M_3 tertiary clusters 1360, and each of the M_3 tertiary clusters 1360 is 3 N (m_3) pieces of load information 1400 belonging to the next cluster 1380 are specified. The N(m_3) pieces of load information 1400 are at least one piece of load information. The tertiary clustering is performed such that the N(m_3) pieces of device-specific load information 1120 included in the N(m_3) pieces of load information 1400 belonging to each tertiary cluster 1380 are similar to each other.

The third-order clustering is performed in the same manner as the second-order clustering except that the similarity 1780 for the third-order clustering is used instead of the similarity 1780 for the second-order clustering.

1.20 Representative Model Acquisition Unit As shown in FIG. 4, the representative model acquisition unit 1726 includes N(m_3) pieces of load information 1400 belonging to each tertiary cluster 1380, which is each of the M_3 tertiary clusters 1360. By generating each representative model 1420 that is representative of, the M_1*M_2*M_3 representative models 1440 are acquired.

When only one piece of load information belongs to each tertiary cluster 1380, the representative model acquisition unit 1726 sets the one piece of load information as it is to each representative model 1420.

On the other hand, when a plurality of load information belongs to each tertiary cluster 1380, the representative model acquisition unit 1726 generates each representative model 1420 from the plurality of load information.

Each representative model 1420 includes total load information and device-specific load information. Each representative model 1420 may include information other than the total load information and the device-specific load information. For example, each representative model 1420 may include temperature information, day of the week information, and the like. The load for each device may be expressed by the ratio of the total load.

The total load information included in each representative model 1420 indicates a temporal change of the total load in one day. Total load is shown hourly.

The device-specific load information included in each representative model 1420 indicates a temporal change of a plurality of device-specific loads in one day. A plurality of device-specific loads are shown every hour.

One day may be replaced by a shorter or longer period than one day. One hour may be replaced by a time interval shorter or longer than one hour. For example, 1 hour may be replaced with 1 minute, 30 minutes, or the like.

1.21 Representative Model Storage Unit The representative model storage unit 1727 stores the M_1*M_2*M_3 representative models 1440 acquired by the representative model acquisition unit 1726.

1.22 Example of Representative Model FIG. 19 is a diagram illustrating an example of a representative model handled by the load estimation device according to the first embodiment with a bar graph and a table. FIG. 20 is a diagram illustrating an example of a representative model handled by the load estimation device according to the first embodiment with a bar graph and a table.

The “representative model DM-A” illustrated in FIG. 19A includes total load information shown by a bar graph. In FIG. 19A, showing the load information for each device included in the “representative model DM-A” is omitted for convenience.

The “representative model DM-A” illustrated in FIG. 19B includes total load information and device-specific information shown in the table.

The “representative model DM-B” illustrated in FIG. 20A includes total load information indicated by a bar graph. In FIG. 20( a ), it is omitted for convenience sake to show the load information by device included in the “representative model DM-B”.

The “representative model DM-B” illustrated in FIG. 20B includes total load information and device-specific information shown in the table.

1.23 Details of Representative Model Acquiring Unit FIG. 21 is a block diagram illustrating a representative model acquiring unit included in the load estimating device according to the first embodiment.

As shown in FIG. 21, the representative model acquisition unit 1726 includes a reference model extraction unit 1880, a correspondence calculation unit 1881 for creating a representative model, and a representative model creation unit 1882.

1.24 Reference Model Extraction Unit The reference model extraction unit 1880 extracts a reference model 1900 for each tertiary cluster 1380 that is each of the M_1*M_2*M_3 tertiary clusters.

When the N(m_3) pieces of load information 1400 belonging to each tertiary cluster 1380 are a plurality of pieces of load information, the reference model extracting unit 1880 selects any one piece of load information from the N(m_3) pieces of load information 1400. Is selected, and the selected load information is extracted as the reference model 1900. When one piece of load information is selected, the similarity between the N(m_3) pieces of load information 1400 may be calculated, and the calculated similarity may be considered.

The extraction method of the reference model 1900 when the N(m_3) pieces of load information 1400 are a plurality of pieces of load information may be changed. For example, an average value of N(m_3) pieces of load information 1400 may be calculated, and the calculated average value may be extracted as the reference model 1900.

1.25 Correspondence Calculation Unit for Creating Representative Model The correspondence calculation unit 1881 for creating a representative model, for each tertiary cluster 1380 that is each of M_1*M_2*M_3 tertiary clusters, for each tertiary cluster 1380 When N(m_3) pieces of load information 1400 belonging to a plurality of pieces of load information are a plurality of pieces of load information, the total load information included in the extracted reference model 1900 and at least one load other than the reference model 1900 belonging to each tertiary cluster 1380 Correspondence relationship information 1940 for creating a representative model showing a time-series correspondence relationship with the total load information included in each of the at least one non-reference model 1920, which is information, is calculated.

The correspondence calculating unit 1881 for creating a representative model indicates the correspondence in time series between the total load information included in the reference model 1900 and the total load information included in each of the at least one non-reference model 1920. When calculating the relationship information, the plurality of time series points of the total load included in the reference model 1900 are approximated to the plurality of time series points of the total load included in each of the at least one non-reference model 1920. A plurality of time series points of the total load included in each of the at least one non-reference model 1920 corresponding to the plurality of time series points of the total load included in the reference model 1900 are specified.

To calculate the time-series correspondence between the total load information included in the reference model 1900 and the total load information included in each of the at least one non-reference model 1920, refer to “1.15 Calculation Method of Correspondence Relationship The calculation method described in "Example" is used. A calculation method other than the calculation method described in “1.15 Example of Calculation Method of Correspondence Relationship” may be used.

1.26 Representative Model Creating Unit The representative model creating unit 1882, for each tertiary cluster 1380 that is each of the M_1*M_2*M_3 tertiary clusters, loads N(m_3) pieces of load information belonging to each tertiary cluster 1380. When 1400 is a plurality of pieces of load information, the representative model 1300 is created by referring to the representative model creating correspondence information 1940 calculated by the representative model creating correspondence calculating unit 1881. As a result, the representative model creation unit 1882 creates M_1*M_2*M_3 representative models 1440 that respectively represent at least one load information belonging to the M_1*M_2*M_3 tertiary clusters.

The time series point d(i) of the load by device included in the representative model at time i is the time series point m(i) of the load by device included in the reference model 1900 at time i and at least one non-reference model 1920. The time series points a(j), b(k),... Corresponding to the time series point m(i) are calculated by Expression (7).

The factors f(m(i),a(j),b(k),...) contained in equation (7) are called m(i),a(j),b(k),... This function returns the representative value of the argument. The representative value is typically an average value. The representative value may be a value obtained by statistical processing other than the average value.

1.27 Example of Creating Representative Model A procedure for creating each representative model 1420 from the reference model 1900 and the non-reference model 1920 when the number of at least one non-reference model 1920 is one reference model will be described.

FIG. 22 is a diagram illustrating an example of a reference model, a non-reference model, and correspondence information handled by the load estimation device according to the first embodiment in the form of a table.

The “reference model MS1” illustrated in FIG. 22A includes the total load information and the device-specific load information shown in the table.

The “non-reference model MXA” illustrated in FIG. 22B includes the total load information and the device-specific load information shown in the table.

The correspondence information illustrated in FIG. 22C indicates a time-series correspondence relationship between the total load information included in the “reference model MS1” and the total load information included in the “non-reference model MXA”.

Correspondence information illustrated in FIG. 22(c) indicates that the equipment included in the "non-reference model MXA" at the time series points m(0) and m(1) of the load by equipment included in the "reference model MS1". It shows that the time series points a(0) and a(2) of different loads correspond to each other. Therefore, the time series points d(0) and d(1) of the load for each device included in the representative model are obtained by the equations (8) and (9), respectively.

When the "non-reference model MXA" does not include the time series point of the load by device corresponding to the time series point m(i) of the load by device included in the "reference model MS1", the time series point m(i) May be set as the time series point d(i) of the load for each device included in the representative model as it is, or the time series point is required by statistical processing from the time series points at times before and after the time when the time series point is required. A time series point at time may be calculated.

By creating each representative model 1420 as described above, each representative model 1420 with no bias is obtained for each tertiary cluster 1380.

1.28 Estimating Unit FIG. 23 is a block diagram illustrating an estimating unit included in the load estimating device according to the first embodiment.

As shown in FIG. 23, the estimation unit 1022 includes a reception unit 1960, a primary cluster selection unit 1961, a representative model selection correspondence calculation unit 1962, a representative model selection similarity calculation unit 1963, and a representative model selection. A unit 1964 and a breakdown calculation unit 1965 are provided.

1.29 Reception Unit The reception unit 1960 acquires the total load information 1080 from the estimation target house 1100.

The reception unit 1960 may acquire the total load information 1080 of the estimation target house 1100 from an operation performed on the screen using an input device such as a keyboard and a mouse, or the total load of the estimation target house 1100 is described. The total load information 1080 may be acquired by reading the file from the outside of the load estimation apparatus 1000, or the total load may be obtained from a database that is an external device of the load estimation apparatus 1000 and stores the total load of the estimation target house 1100. The total load information 1080 may be acquired by receiving.

1.30 Example of Total Load Information FIG. 24 is a table illustrating an example of the total load information of the estimation target house handled by the load estimation device of the first embodiment.

The total load information illustrated in FIG. 24 is simple time series data including a plurality of times and a plurality of total loads respectively corresponding to the plurality of times.

1.31 Primary cluster selection unit The primary cluster selection unit 1961 is total load information 1660 to which the total load information 1080 of the estimation target house 1100 is given, and the feature amount 1081 of the estimation target house 1100 is given. When it is 1661, the selected primary cluster 1460 identified by the primary cluster number derived by the derivation formula 1280 is selected from the M_1 primary clusters 1220.

FIG. 25 is a diagram illustrating an example of information extracted from the total load information and the feature amount of the estimation target house handled by the load estimation device of the first embodiment, and derivation in the load estimation device of the first embodiment. It is a figure containing an explanatory view explaining an example of derivation of a primary cluster number by a formula.

In the example illustrated in FIG. 25A, the information extracted from the total load information 1080 of the estimation target house 1100 is the average value and the maximum value for one hour. Further, the information extracted from the feature amount 1081 of the estimation target house 1100 is the floor area and the number of residents.

In the example illustrated in FIG. 25(b), the nodes 1702 and 1703 determine the magnitude of the maximum value and average value for one hour and the reference value, respectively, and the nodes 1700 and 1701 respectively determine the floor area and the number of residents. And the reference value are compared. As a result, the derived primary cluster number 1711 is selected from the primary cluster number candidates 1710, 1711, 1712, 1713, and 1714.

1.32 Correspondence Calculation Unit for Representative Model Selection The correspondence relationship calculation unit 1962 for representative model selection selects the selected primary cluster 1460 from the M_1*M_2*M_3 representative models 1440 stored in the representative model storage unit 1727. Select two or more representative models 1500 representing the load information 1480 belonging to. In addition, the correspondence calculation unit 1962 for selecting a representative model is included in each representative model that is the total load information 1080 of the estimation target house 1100 acquired by the reception unit 1960 and each of the selected two or more representative models 1500. Correspondence relationship information 1980 for selecting a representative model, which indicates a time series correspondence relationship with the total load information, is calculated.

When calculating the correspondence information 1980 for selecting the representative model, the correspondence calculating unit 1962 for selecting the representative model determines that the plurality of time series points of the total load included in the total load information 1080 of the estimation target house 1100 are A plurality of total loads included in each representative model corresponding to the plurality of time series points of the total load included in the total load information 1080 are approximated to the plurality of time series points of the total load included in the representative model. Identify time series points.

For calculation of the time series correspondence between the total load information 1080 of the estimation target house 1100 and the total load information included in each representative model, the calculation described in "1.15 Example of Correspondence Calculation Method" is performed. Method is used. A calculation method other than the calculation method described in “1.15 Example of Calculation Method of Correspondence Relationship” may be used.

1.33 Representative Model Selection Similarity Calculation Unit The representative model selection similarity calculation unit 1963 selects the selected primary cluster 1460 from the M_1*M_2*M_3 representative models 1440 stored in the representative model storage unit 1727. Select two or more representative models 1500 representing the load information 1480 belonging to. The representative model selection similarity calculation unit 1963 is included in each representative model that is each of the total load information 1080 of the estimation target house 1100 acquired by the reception unit 1960 and each of the selected two or more representative models 1500. The similarity 1990 for selecting the representative model with the total load information is calculated.

The similarity calculation unit 1963 for representative model selection uses the correspondence information 1980 for representative model selection calculated by the correspondence calculation unit 1962 for representative model selection when calculating the similarity 1990 for representative model selection. refer.

The representative model selection similarity calculation unit 1963 refers to the representative model selection correspondence information 1980 to respectively correspond to a plurality of time series points of the total load included in the total load information 1080 of the estimation target house 1100. Then, a plurality of time series points of the total load included in each representative model are specified.

In addition, the similarity calculation unit 1963 for selecting a representative model includes a plurality of time series points of total loads included in the total load information 1080 of the estimation target house 1100 and a plurality of time series of total loads included in each representative model. The degree of similarity between and is set to the degree of similarity 1990 for selecting a representative model.

The calculation method described in “1.15 Example of Calculation Method of Correspondence Relationship” is used, and a plurality of time series points of the total load included in the total load information 1080 of the estimation target house 1100 and the total load included in each representative model are used. When the cost required to correspond to each of the plurality of time series is used as the similarity 1990 for selecting the representative model, the similarity calculating unit 1963 for selecting the representative model is omitted, and the similarity calculating unit 1963 for selecting the representative model is omitted. The correspondence calculation unit 1962 also serves as the similarity calculation unit for selecting the representative model.

1.34 Representative Model Selection Unit The representative model selection unit 1964 determines M_1*M_2*M_3 representative models 1440 based on the similarity 1990 for representative model selection calculated by the similarity calculation unit 1963 for representative model selection. One selected representative model 1520 is selected.

When selecting the selected representative model 1520, the representative model selection unit 1964 determines that the similarity between the total load information 1080 of the estimation target house 1100 acquired by the reception unit 1960 and the total load information included in the candidate representative model is A selection criterion is used such that the higher the value, the easier it is to select a candidate representative model.

As the selection criterion, the selection representative model 1520 is typically selected such that the similarity between the total load information 1080 of the estimation target house 1100 and the total load information included in the selection representative model 1520 is the highest similarity. The standard is to do so. However, when the selection criterion includes consideration elements other than the similarity 1990 for selecting the representative model and the selection representative model 1520 is selected, in addition to the similarity 1990 for selecting the representative model, a similarity other than the similarity 1990 for selecting the representative model is used. Consideration factors may be considered. For example, the selection criterion includes M_1*M_2*M_3 pieces of temperature information, M_1*M_2*M_3 pieces of day-of-week information, etc., which are associated with M_1*M_2*M_3 representative models 1440, and the selection representative model 1520 is included. When selecting, M_1*M_2*M_3 temperature information, M_1*M_2*M_3 day information, etc. may be considered in addition to the similarity 1990 for representative model selection.

1.35 Breakdown Calculation Unit The breakdown calculation unit 1965 uses the total load information 1080 of the estimation target house 1100 acquired by the reception unit 1960, the total load information included in the selected representative model 1520, and the load information for each device of the estimation target house 1100. The breakdown of the total load information 1080 is calculated. As a result, the device-specific load information 1120 of the estimation target house 1100 indicating the breakdown of the total load information 1080 is calculated. In the calculation of the breakdown, the load information by device included in the selected representative model 1520 indicating the breakdown of the total load information included in the selected representative model 1520 is reflected in the load information by device 1120 indicating the breakdown of the total load information 1080. Is done like.

The breakdown calculation unit 1965 is included in the total load information 1080 of the estimation target house 1100 and the selected representative model 1520, which are included in the representative model selection correspondence information 1980 calculated by the representative model selection correspondence relation calculation unit 1962. The total load information 1080 and the total load information included in the selected representative model 1520 are associated in time series on the basis of the correspondence information for breakdown calculation showing the corresponding relationship in time series with the total load information, The breakdown of the total load information 1080 is calculated from the total load information 1080 and the total load information and the device-specific load information included in the selected representative model 1520.

When the total load time series points X=x_t(1), x_t(2),..., x_t(m) included in the total load information 1080 of the estimation target house 1100 and the total load included in the selected representative model 1520 Correspondence relation information for breakdown calculation that the time series point y_t(j) corresponds to the time series point x_t(i) at the series point Y=y_t(1),y_t(2),...,y_t(n) , The time series point E(X,t(i)) of the load by device at the time i of the device EP of the estimation target house 1100 is the time series point A(X,X,t of the total load of the estimation target house 1100. t(i)), the time series point A(Y,t(i)) of the total load included in the selected representative model 1520, and the time series point E(Y of the load of each device EP included in the selected representative model 1520. , t(j)), and is calculated by the equation (10).

The time series point E(X,t(i)) may be obtained by an equation other than the equation (10).

When the time interval indicating the total load in the total load information 1080 of the estimation target house 1100 is different from the time interval indicating the total load in the total load information included in the selected representative model 1520, the total load indicating the total load in a short time interval. After the process of aligning the information with the total load information indicating the total load at a long time interval, the time series point E(X,t(i)) is obtained as described above.

According to the calculation of the breakdown based on the correspondence information for the breakdown calculation, the time change of the total load included in the selected representative model 1520 is similar to the time change of the total load included in the total load information 1080 of the estimation target house 1100. When the total load occurrence time in the selected representative model 1520 is different from the total load occurrence time in the total load information 1080, the time when the total load in the selected representative model 1520 becomes higher is the time when the total load in the total load information 1080 becomes higher. Even if the difference is, the matching for eliminating the difference is performed and the device-specific load information 1120 of the estimation target house 1100 is accurately estimated.

1.36 Advantages of Using Correspondence Relationship Information for Breakdown Calculation FIG. 26 is a diagram illustrating, by a line graph, an example of total load information to be estimated which is handled by the load estimating apparatus according to the first embodiment.

In FIG. 26, "total load information TL" is illustrated.

FIG. 27 is a diagram illustrating an example of a selected representative model handled by the load estimation device according to the first embodiment by a stacked bar graph.

The “selected representative model SM1” illustrated in FIG. 27 includes total load information indicated by the entire plurality of components in the stacked bar graph and load information by device indicated by the plurality of components in the stacked bar graph.

FIG. 28 is a diagram illustrating an example of the total load information of the estimation target and the selected representative model handled by the load estimation device of the first embodiment, which are overlapped and illustrated so as to be comparable.

In FIG. 28, "total load information TL" is illustrated. Further, the “selected representative model SM1” illustrated in FIG. 28 includes total load information indicated by the entire plurality of components in the stacked bar graph and device-specific load information indicated by the plurality of components in the stacked bar graph.

The total load information included in the “total load information TL” and the “selected representative model SM1” has a common feature that the amount of power consumed in the morning, daytime, and evening is large. Therefore, the “selected representative model SM1” is selected as the representative model including the total load information similar to the “total load information TL”. However, the evening time when the amount of power consumed in the “total load information TL” becomes large deviates from the evening time when the amount of power consumed in the total load information included in the “selected representative model SM1” becomes large. ..

In the following, for each of the case where the correspondence information for breakdown calculation is not used and the correspondence information for breakdown calculation is used, the estimation target housing 1100 from the “total load information TL” and the “selected representative model SM1” is calculated. The result of the estimation of the load information for each device 1120 will be described.

(A) Case where Correspondence Relationship Information for Breakdown Calculation is Not Used FIG. 29 is an example of correspondence relationship information used when correspondence relationship information for breakdown calculation handled by the load estimation apparatus of the first embodiment is not used. It is a figure illustrated by a table.

FIG. 30 is a diagram illustrating, by a stacked bar graph, the load information for each device estimated when the correspondence information for selecting the representative model handled in the load estimation device of the first embodiment is not used.

In the “device-specific load information SL0” illustrated in FIG. 30, the device-specific load information is indicated by a plurality of components in the stacked bar graph.

According to the correspondence information illustrated in FIG. 29, the plurality of time series points included in the “total load information TL” include the plurality of time series points included in the “selected representative model SM1” and acquired at the same time. Corresponding to each. Therefore, the features of the total load information and the device-specific load information included in the “total load information TL” and the “selected representative model SM1” are not considered.

When the correspondence information illustrated in FIG. 29 is used, the breakdown indicated by the “selected representative model SM1” at 21:00 when the total load included in the “selected representative model SM1” is low is “total load information TL”. Is used to calculate the breakdown of “total load information TL” at 21:00. Therefore, in the “device-specific load information SL0” illustrated in FIG. 30, the ratio of the load of the living room to the total load at 21:00 is high, reflecting the details shown in the “selected representative model SM1” at 21:00. Has become.

However, it is appropriate that the ratio of the living load to the total load at 21:00 is obtained by reflecting the breakdown shown by the “selected representative model SM1” at 18:00 when the total load included in the “selected representative model SM1” becomes high. Therefore, the “device-specific load information SL0” is likely to include a large error.

(B) Case of Utilizing Correspondence Information for Breakdown Calculation FIG. 31 is a diagram illustrating an example of correspondence information for breakdown calculation handled by the load estimation apparatus according to the first embodiment.

FIG. 32 is a diagram illustrating, by a stacked bar graph, the load information for each device estimated when the correspondence information for breakdown calculation used in the load estimation device according to the first embodiment is used.

In the “device-specific load information SL1” illustrated in FIG. 32, the device-specific load information is indicated by a plurality of components in the stacked bar graph.

When the correspondence information illustrated in FIG. 31 is used, the breakdown indicated by the “selected representative model SM1” at 15:00 when the total load included in the “selected representative model SM1” is low is “total load information TL”. It is used to calculate the breakdown of the “total load information TL” at 15:00-18:00 when the value is low. Further, the breakdown indicated by the "selected representative model SM1" at 18:00 when the total load included in the "selected representative model SM1" is high is the breakdown of the "total load information TL" at 21:00 when the "total load information TL" is high. Used for calculation. Therefore, in the “load information SL1 for each device” illustrated in FIG. 32, the ratio of the load of the living room to the total load at 21:00 is appropriate, reflecting the details shown by the “selected representative model SM1” at 18:00. It has become. Therefore, the “device-specific load information SL1” is unlikely to include a large error.

1.37 Advantages of classifying load information using feature amount Generally, even if the total load information of the first house is similar to the total load information of the second house, the first house The device-specific load information of (2) is not always similar to the device-specific load information of the second house. For example, when the resident of the first house is one resident and the resident of the second house is two or more residents, the load information by device of the first house is the second resident. In many cases, it is not similar to the load information for each device in a house. This means that one resident often has one air conditioner, but two or more residents often have two or more air conditioners. You can understand.

Therefore, when the selection representative model 1520 is selected without using the N feature amounts 1041 of the monitoring house 1060 and the feature amount 1081 of the estimation target house 1100, it is similar to the load information 1120 by device of the estimation target house 1100. The selected representative model 1520 including the load information for each device may be selected. Therefore, the load information for each device 1120 may not be accurately estimated.

On the other hand, by using the N feature amounts 1041 of the monitoring house 1060 and the feature amount 1081 of the estimation target house 1100, the load information of the monitoring house having the feature amount similar to the feature amount 1081 of the estimation target house 1100 is obtained. If the selected primary cluster 1460 to which the selected selected primary cluster 1460 belongs is selected and the selected representative model 1520 is selected from the two or more representative models 1500 representing the load information belonging to the selected selected primary cluster 1460, the device of the estimation target house 1100 The selected representative model 1520 including the load information for each device similar to the load information for another 1120 is selected. Therefore, the load information for each device 1120 is accurately estimated.

1.38 Advantages of Load Estimating Device According to First Embodiment According to the load estimating device 1000 according to the first embodiment, since it is not necessary to acquire the load for each device from the estimation target house 1100, a measuring device that measures the load for each device Does not need to be installed in the estimation target house 1100, and the load information 1120 by device of the estimation target house 1100 can be easily obtained.

Generally speaking, the characteristics of a house affect the load by device, which is a breakdown of the total load of the house. Therefore, when the temporal change of the total load of the estimation target house 1100 is similar to the temporal change of the total load of the monitoring house and the characteristics of the estimation target house 1100 are similar to the characteristics of the monitoring house, the total load of the estimation target house 1100 is The load by device, which is a breakdown of the load, is likely to be similar to the load by device, which is a breakdown of the total load of the monitoring house. Then, according to the load estimation apparatus 1000 of the first embodiment, the estimation target house is calculated from the two or more representative models 1500 that represent the load information of the monitoring house having the feature amount similar to the feature amount 1081 of the estimation target house 1100. The selected representative model 1520 having total load information similar to the total load information 1080 of 1100 is selected. Further, the load information 1120 for each device of the estimation target house 1100 is estimated using the selection representative model 1520. As a result, the device-specific load information 1120 is accurately estimated.

Furthermore, according to the load estimation apparatus 1000 of the first embodiment, when the selection representative model 1520 is selected, the total load information 1080 of the estimation target house 1100 and the total load information included in the candidate representative model are time-series. Since the similarity between the total load information 1080 and the total load information included in the candidate representative model is calculated after the above correspondence, an appropriate selected representative model 1520 is selected. As a result, the device-specific load information 1120 is estimated with higher accuracy.

Second Embodiment
2.1 Differences between Embodiment 1 and Embodiment 2 FIG. 33 is a block diagram illustrating a load estimation device according to Embodiment 2.

In the load estimation apparatus 1000 according to the first embodiment, N pieces of load information 1040 and N pieces of feature amount 1041 are classified into M_1 primary clusters 1220. Further, a derivation formula 1280 for deriving the primary cluster number 1662 from the given total load information 1660 and the feature amount 1661 is created. Further, when the total load information 1080 and the feature amount 1081 of the estimation target house 1100 are given, the selected primary cluster 1460 identified by the primary cluster number derived by the derivation formula 1280 is selected from the M_1 primary clusters 1220. Selected. Further, the selected representative model 1520 is selected from the two or more representative models 1500 representing the load information belonging to the selected primary cluster 1460.

On the other hand, in the load estimation apparatus 2000 according to the second embodiment, N pieces of load information 1040, N pieces of characteristic amount 1041 and N pieces of temperature information 1042 are classified into M_1 primary clusters 1220. Further, a derivation formula 1280 for deriving the primary cluster number 1662 from the given total load information 1660, feature amount 1661 and temperature information is created. Further, when the total load information 1080, the feature amount 1081 and the temperature information 1082 of the estimation target house 1100 are given, the selected primary cluster 1460 identified by the primary cluster number derived by the derivation formula 1280 is M_1 1 It is selected from the next cluster 1220.

The configuration of the load estimation device 2000 according to the second embodiment that causes the above differences will be described below. The load estimating apparatus 2000 according to the second embodiment may be used as it is or after the configuration of the load estimating apparatus according to another embodiment is modified as long as the adoption of the configuration that causes the above difference is not hindered.

2.2 Load Estimation Device The load estimation device 2000 illustrated in FIG. 33 includes a load ratio classification unit 1020, a representative model preparation unit 1021 and an estimation unit 1022.

The load estimation device 2000 acquires N pieces of load information 1040 and N pieces of characteristic amount 1041 from the monitoring house 1060, and acquires total load information 1080 and a characteristic amount 1081 from the estimation target house 1100. In addition, the load estimation device 2000 acquires N pieces of temperature information 1042 from the temperature information source 1061 and acquires the temperature information 1082 from the temperature information source 1101.

2.3 Temperature information handled by the load estimation device
The N pieces of temperature information 1042 include temperature information associated with each piece of load information 1160. Therefore, the N pieces of temperature information 1042 are associated with the N pieces of load information 1040, respectively.

The temperature information associated with each load information 1160 indicates the temperature of the area where the monitoring house from which the load information 1160 is acquired consumes electric energy when the monitoring house from which the load information 1160 is acquired consumes electrical energy. Since each load information 1160 is the load information for one day, the temperature information indicating the temperature when the total load and the device-specific load included in each load information 1160 associated with each load information 1160 are measured is also one day. Minute temperature information.

The temperature information 1082 indicates the temperature of the area where the estimation target house 1100 is installed when the estimation target house 1100 consumes electric energy. Since the total load information 1080 is the total load information for one day, the temperature information 1082 indicating the temperature when the total load included in the total load information 1080 is measured is also the temperature information for one day.

The temperature information source 1061 may be a site that publishes the temperature information on the Internet, may be a device installed in the monitoring house 1060 to which the temperature information is manually input, or the monitoring house 1060. It may be an installed thermometer.

The temperature information source 1101 may also be a site that publishes temperature information on the Internet, or may be a device installed in the estimation target house 1100 to which the temperature information is manually input, or an estimation target house. It may be a thermometer installed at 1100. The temperature information source 1101 may be the same temperature information source as the temperature information source 1061.

Each of the N pieces of temperature information 1042 may be time-series data including the temperature at each time, or may be data extracted from the time-series data such as the maximum temperature and the minimum temperature, or the average. It may be data obtained by processing the time series data such as temperature.

The temperature information 1082 may be time-series data including the temperature at each time, may be data extracted from the time-series data such as the maximum temperature and the minimum temperature, or may be the average time and the like. It may be data obtained by processing the series data.

2.4 Load Ratio Classifying Unit FIG. 34 is a block diagram illustrating a load ratio classifying unit included in the load estimating device according to the second embodiment.

The load ratio classification unit 1020 includes a load information storage unit 1540, a device-based load ratio calculation unit 1541, a primary cluster number determination unit 1542, a learning data classification unit 1543, and a derivation formula creation unit 1544, and further includes an air temperature information acquisition unit 1545. Prepare

2.5 Temperature Information Acquisition Unit The temperature information acquisition unit 1545 acquires N pieces of temperature information 1042 from the temperature information source 1061.

2.6 Primary Clustering Unit The primary clustering unit 1581 has the monitoring house stored in the load information storage unit 1540 according to the primary clustering similarity 1600 calculated by the primary clustering similarity calculating unit 1580. 1060 N load information 1040 and N feature amounts 1041 and N temperature information 1042 acquired by the temperature information acquisition unit 1545 are classified into M_1 primary clusters 1220, and M_1 primary clusters 1220. N(m_1) pieces of load information 1260, N(m_1) pieces of characteristic amount 1261 and N(m_1) pieces of temperature information belonging to each primary cluster 1240 which is each of 1220 are specified. The N(m_1) feature amounts 1261 are associated with the N(m_1) load information 1260, respectively. The N(m_1) pieces of temperature information are associated with the N(m_1) pieces of load information 1260, respectively. The N(m_1) pieces of load information 1260 are at least one piece of load information. The N(m_1) feature quantities 1261 are at least one feature quantity. The N(m_1) pieces of temperature information are at least one piece of temperature information.

2.7 Derivation Formula Creation Unit The derivation formula creation unit 1544 collects N(m_1) load information 1260, N(m_1) feature amounts 1261 and N(m_1) temperature information belonging to each primary cluster 1240. After learning, a derivation formula 1280 for deriving the primary cluster number 1662 from the given total load information 1660, the given feature amount 1661 and the given temperature information is created. The derived primary cluster number 1662 identifies the primary cluster. Therefore, the derived derivation formula 1280 derives a primary cluster from the given total load information 1660, the given feature amount 1661, and the given temperature information.

The derivation formula 1280 is created by learning N(m_1) pieces of load information 1260, N(m_1) pieces of feature amount 1261 and N(m_1) pieces of temperature information belonging to each primary cluster 1240. Therefore, the primary cluster number 1662 derived by the derivation formula 1280 is the primary cluster to which the total load information 1660, the characteristic amount 1661, and the temperature information that are most similar to the given total load information 1606, the characteristic amount, and the temperature information belong. This is the primary cluster number to be identified. It is the total load information 1660 to which the total load information included in the load information belonging to the specific primary cluster is given, and the characteristic amount 1661 to which the characteristic amount associated with the load information is given. When the temperature information belonging to the cluster is the given temperature information, the primary cluster number derived by the derivation formula 1280 is the primary cluster number for identifying the specific primary cluster.

2.8 Example of Derivation of Primary Cluster Number FIG. 35 is an explanatory diagram illustrating an example of derivation of a primary cluster number by a derivation formula performed in the load estimation device of the second embodiment.

In the derivation of the primary cluster number by the derivation formula 1280, the given total load information 1660, the given feature amount 1661 and the given temperature information are analyzed, and the information that influences the derived primary cluster number 1662. As shown in FIG. 35, the nodes 1700, 1701, 1703, and 1705 determine whether the extracted information and the reference value are large or small, whether the extracted information meets the conditions, and the like. Then, based on the result, the derived primary cluster number 1662 is selected from the primary cluster number candidates 1710, 1711, 1712, 1713 and 1714. The extracted information may be information directly extracted from the total load information 1660, the characteristic amount 1661 and the temperature information, or may be extracted by processing the total load information 1660, the characteristic amount 1661 and the temperature information. The information may be The information extracted by processing the total load information 1660 is included in the total load information 1660 including the total load statistical values such as the maximum value, the minimum value, and the average value of the total load included in the total load information 1660 for one hour. It is a numerical value or the like obtained by performing a Fourier transform on the total load. The information extracted by processing the temperature information may be a representative value of the temperature such as the maximum temperature that is a determination target in the node 1705, or a statistical value of the temperature such as the average temperature. Only part of this information may be used.

2.9 Estimating Unit FIG. 36 is a block diagram illustrating an estimating unit included in the load estimating device according to the second embodiment.

As shown in FIG. 36, the estimation unit 1022 includes a reception unit 1960, a primary cluster selection unit 1961, a correspondence calculation unit 1962 for representative model selection, a similarity calculation unit 1963 for representative model selection, and a representative model selection. A unit 1964 and a breakdown calculation unit 1965 are provided, and an air temperature information acquisition unit 1966 is further provided.

2.10 Temperature Information Acquisition Unit The temperature information acquisition unit 1966 acquires the temperature information 1082 from the temperature information source 1101.

2.11 Primary Cluster Selection Unit The primary cluster selection unit 1961 is total load information 1660 given the total load information 1080 of the estimation target house 1100, and the characteristic amount 1081 of the estimation target house 1100. 1661, the selected primary cluster 1460 identified by the primary cluster number derived by the derivation formula 1280 when the temperature information 1082 is the given temperature information is selected from the M_1 primary clusters 1220.

FIG. 37 is a diagram illustrating an example of information extracted from the total load information and the characteristic amount of the estimation target house handled by the load estimation device of the second embodiment and the temperature information, and the load of the second embodiment. It is a figure including an explanatory view explaining an example of derivation of a primary cluster number by a derivation formula in an estimating device.

In the example illustrated in FIG. 37(a), the information extracted from the total load information 1080 of the estimation target house 1100 is the average value and the maximum value for one hour. Further, the information extracted from the feature amount 1081 of the estimation target house 1100 is the floor area and the number of residents. The information extracted from the temperature information 1082 is the maximum temperature.

In the example illustrated in FIG. 37(b), the node 1703 determines the magnitude of the one-hour average value and the reference value, and the nodes 1700 and 1701 respectively determine the floor area and the number of residents and the reference value. Is determined, and the size of the maximum temperature and the reference value is determined in the node 1705. As a result, the derived primary cluster number 1714 is selected from the primary cluster number candidates 1710, 1711, 1712, 1713, and 1714.

2.12 Advantages of Load Estimating Apparatus According to Second Embodiment The load estimating apparatus 2000 according to the second embodiment has the same advantages as the advantages of the load estimating apparatus 1000 according to the first embodiment.

Further, generally, the characteristics of the house and the temperature when the house consumes electric energy affect the load by device, which is a breakdown of the total load of the house. For this reason, the time change of the total load of the estimation target house 1100 is similar to the time change of the total load of the monitoring house, the characteristics of the estimation target house 1100 are similar to the characteristics of the monitoring house, and the estimation target house 1100 outputs electric energy. When the temperature when consumed is similar to the temperature when the monitoring house consumes electric energy, the load by device, which is a breakdown of the total load of the estimated target house 1100, is a device that is a breakdown of the total load of the monitoring house. It is likely to be similar to another load. Then, according to the load estimation device 2000 of the second embodiment, the estimation target house 1100 has the characteristic amount similar to the characteristic amount 1081 of the estimation target house 1100, and the temperature information when the electric energy is consumed indicates that the estimation target house 1100 indicates the electric energy. The selected representative model 1520 having the total load information similar to the total load information 1080 of the estimation target house 1100 is selected from the two or more representative models 1500 representing the load information of the monitoring house similar to the temperature information 1082 at the time of consumption. To be done. Further, the load information 1120 for each device of the estimation target house 1100 is estimated using the selection representative model 1520. As a result, the device-specific load information 1120 is accurately estimated.

Third Embodiment
3.1 Differences between First Embodiment and Third Embodiment FIG. 38 is a block diagram illustrating a load estimation device according to the third embodiment.

In the load estimation apparatus 1000 according to the first embodiment, N pieces of load information 1040 and N pieces of feature amount 1041 are classified into M_1 primary clusters 1220. Further, a derivation formula 1280 for deriving the primary cluster number 1662 from the given total load information 1660 and the feature amount 1661 is created. Further, when the total load information 1080 and the feature amount 1081 of the estimation target house 1100 are given, the selected primary cluster 1460 identified by the primary cluster number derived by the derivation formula 1280 is selected from the M_1 primary clusters 1220. Selected. Further, the selected representative model 1520 is selected from the two or more representative models 1500 representing the load information belonging to the selected primary cluster 1460.

On the other hand, in the load estimation device 3000 according to the third embodiment, the N load information 1040, the N feature amounts 1041 and the N weather information 1043 are classified into M_1 primary clusters 1220. Further, a derivation formula 1280 for deriving the primary cluster number 1662 from the given total load information 1660, feature amount 1661 and weather information is created. Further, when the total load information 1080, the feature amount 1081, and the weather information 1083 of the estimation target house 1100 are given, the selected primary cluster 1460 identified by the primary cluster number derived by the derivation formula 1280 is M_1 1 It is selected from the next cluster 1220.

Hereinafter, the configuration of the load estimation device 3000 according to the third embodiment that causes the above differences will be described. The load estimating device 3000 according to the third embodiment may be used as it is or after the structure of the load estimating device according to another embodiment is modified as long as it does not hinder the adoption of the structure that causes the difference.

3.2 Load Estimating Device A load estimating device 3000 illustrated in FIG. 38 includes a load ratio classifying unit 1020, a representative model preparing unit 1021 and an estimating unit 1022.

The load estimation device 3000 acquires N load information 1040 and N feature amounts 1041 from the monitoring house 1060, and acquires total load information 1080 and feature amount 1081 from the estimation target house 1100. In addition, the load estimation device 3000 acquires N pieces of weather information 1043 from the weather information source 1062 and acquires the weather information 1083 from the weather information source 1102.

3.3 Weather information handled by the load estimation device
The N pieces of weather information 1043 include weather information associated with each load information 1160. Therefore, the N pieces of weather information 1043 are associated with the N pieces of load information 1040, respectively.

The weather information associated with each load information 1160 indicates the weather in the area where the monitoring house from which the load information 1160 is acquired consumes electric energy when the monitoring house from which the load information 1160 is acquired consumes electric energy. The weather is represented by "clear", "cloudy", "sometimes cloudy", and the like. Since each load information 1160 is the load information for one day, the weather information indicating the weather when the total load and the device-specific load included in each load information 1160 associated with each load information 1160 are measured is also one day. Minute weather information.

The weather information 1083 indicates the weather in the area where the estimation target house 1100 is installed when the estimation target house 1100 consumes electric energy. Since the total load information 1080 is the total load information for one day, the weather information 1083 indicating the weather when the total load included in the total load information 1080 is also the weather information for one day.

The weather information source 1062 may be a site that publishes weather information on the Internet, may be a device installed in the monitoring house 1060 to which the weather information is manually input, or the monitoring house 1060. It may be an installed camera.

The weather information source 1102 may also be a site that publishes weather information on the Internet, or may be a device installed in the estimation target house 1100 to which the weather information is manually input, or an estimation target house. It may be a camera installed at 1100. The weather information source 1102 may be the same weather information source as the weather information source 1062.

Each of the N pieces of weather information 1043 may be time-series data including the weather at each time, may be data extracted from the time-series data, or may process the time-series data. It may be data obtained by.

The weather information 1083 may be time series data including the weather at each time, may be data extracted from the time series data, or may be data obtained by processing the time series data. It may be.

3.4 Load Ratio Classifying Unit FIG. 39 is a block diagram illustrating a load ratio classifying unit included in the load estimating device according to the third embodiment.

The load ratio classification unit 1020 includes a load information storage unit 1540, a device-based load ratio calculation unit 1541, a primary cluster number determination unit 1542, a learning data classification unit 1543, and a derivation formula creation unit 1544, and further includes a weather information acquisition unit 1546. Prepare

3.5 Weather Information Acquisition Unit The weather information acquisition unit 1546 acquires N pieces of weather information 1043 from the weather information source 1062.

3.6 Primary Clustering Unit The primary clustering unit 1581 monitors the monitoring houses stored in the load information storage unit 1540 according to the primary clustering similarity 1600 calculated by the primary clustering similarity calculating unit 1580. 1060 N load information 1040 and N feature amounts 1041, and N weather information 1043 acquired by the weather information acquisition unit 1546 are classified into M_1 primary clusters 1220, and M_1 primary clusters 1220. The load information 1260 of N(m_1) pieces, the feature amount 1261 of N(m_1) pieces and the weather information of N(m_1) pieces belonging to each primary cluster 1240 which is each of 1220 are specified. The N(m_1) feature amounts 1261 are associated with the N(m_1) load information 1260, respectively. The N(m_1) pieces of weather information are associated with the N(m_1) pieces of load information 1260, respectively. The N(m_1) pieces of load information 1260 are at least one piece of load information. The N(m_1) feature quantities 1261 are at least one feature quantity. The N(m_1) pieces of weather information are at least one piece of weather information.

3.7 Derivation Formula Creation Unit The derivation formula creation unit 1544 collects N(m_1) load information 1260, N(m_1) feature quantities 1261 and N(m_1) weather information belonging to each primary cluster 1240. After learning, a derivation formula 1280 for deriving the primary cluster number 1662 from the given total load information 1660, the given feature amount 1661 and the given weather information is created. The derived primary cluster number 1662 identifies the primary cluster. Therefore, the derived derivation formula 1280 derives a primary cluster from the given total load information 1660, the given feature amount 1661, and the given weather information.

The derivation formula 1280 is created by learning N(m_1) pieces of load information 1260, N(m_1) pieces of feature amount 1261 and N(m_1) pieces of weather information belonging to each primary cluster 1240. Therefore, the primary cluster number 1662 derived by the derivation formula 1280 is the primary cluster to which the total load information 1660, the feature amount 1661, and the weather information that are most similar to the given total load information, the feature amount, and the weather information belong. This is the primary cluster number to be identified. It is the total load information 1660 to which the total load information included in the load information belonging to the specific primary cluster is given, and the characteristic amount 1661 to which the characteristic amount associated with the load information is given. When the weather information belonging to the cluster is given weather information, the primary cluster number derived by the derivation formula 1280 is the primary cluster number for identifying the particular primary cluster.

3.8 Derivation Example of Primary Cluster Number FIG. 40 is an explanatory diagram illustrating an example of derivation of a primary cluster number by a derivation formula performed in the load estimation device of the third embodiment.

In the derivation of the primary cluster number by the derivation formula 1280, the given total load information 1660, the given feature amount 1661 and the given weather information are analyzed, and the information that influences the derived primary cluster number 1662. As shown in FIG. 40, the nodes 1700, 1701, 1703, and 1706 determine the magnitude of the extracted information and the reference value, whether the extracted information meets the conditions, and the like. Then, based on the result, the derived primary cluster number 1662 is selected from the primary cluster number candidates 1710, 1711, 1712, 1713 and 1714. The extracted information may be information directly extracted from the total load information 1660, the feature amount 1661 and the weather information, or may be extracted by processing the total load information 1660, the feature amount 1661 and the weather information. The information may be The information extracted by processing the total load information 1660 is included in the total load information 1660 including the total load statistical values such as the maximum value, the minimum value, and the average value of the total load included in the total load information 1660 for one hour. It is a numerical value or the like obtained by performing a Fourier transform on the total load. Only part of this information may be used.

3.9 Estimating Unit FIG. 41 is a block diagram illustrating an estimating unit included in the load estimating device according to the third embodiment.

As shown in FIG. 41, the estimation unit 1022 includes a reception unit 1960, a primary cluster selection unit 1961, a correspondence calculation unit 1962 for representative model selection, a similarity calculation unit 1963 for representative model selection, and a representative model selection. A unit 1964 and a breakdown calculation unit 1965 are provided, and a weather information acquisition unit 1967 is further provided.

3.10 Weather Information Acquisition Unit The weather information acquisition unit 1967 acquires the weather information 1083 from the weather information source 1102.

3.11 Primary Cluster Selection Unit The primary cluster selection unit 1961 is total load information 1660 to which total load information 1080 of the estimation target house 1100 is given, and feature amount 1081 of the estimation target house 1100 is given. 1661, the selected primary cluster 1460 identified by the primary cluster number 1662 derived by the derivation formula 1280 when the weather information 1083 is given weather information is selected from the M_1 primary clusters 1220.

FIG. 42 is a diagram illustrating an example of information extracted from the total load information and the feature amount of the estimation target house handled by the load estimation device of the third embodiment and the weather information, and the load of the third embodiment. It is a figure including an explanatory view explaining an example of derivation of a primary cluster number by a derivation formula in an estimating device.

In the example illustrated in FIG. 42(a), the information extracted from the total load information 1080 of the estimation target house 1100 is the average value and the maximum value for one hour. Further, the information extracted from the feature amount 1081 of the estimation target house 1100 is the floor area and the number of residents. The information extracted from the weather information 1083 is the weather.

In the example shown in FIG. 42(b), the node 1703 determines the magnitude of the one-hour average value and the reference value, and the nodes 1700 and 1701 respectively determine the floor area and the number of residents and the reference value. Is determined, and it is determined at node 1706 that the weather condition is not met. As a result, the derived primary cluster number 1713 is selected from the primary cluster number candidates 1710, 1711, 1712, 1713, and 1714.

3.12 Advantages of Load Estimating Device According to Second Embodiment The load estimating device 3000 according to the third embodiment has the same advantages as those of the load estimating device 1000 according to the first embodiment.

Further, generally, the characteristics of the house and the weather when the house consumes electric energy affect the load by device, which is a breakdown of the total load of the house. For this reason, the time change of the total load of the estimation target house 1100 is similar to the time change of the total load of the monitoring house, the characteristics of the estimation target house 1100 are similar to the characteristics of the monitoring house, and the estimation target house 1100 outputs electric energy. When the weather when consumed is similar to the weather when the monitoring house consumes electric energy, the device-specific load, which is a breakdown of the total load of the estimation target house 1100, is a device that is a breakdown of the total load of the monitoring house. It is likely to be similar to another load. Then, according to the load estimation apparatus 3000 of the third embodiment, the estimated target house 1100 has the characteristic amount similar to the characteristic amount 1081 of the estimated target house 1100, and the weather information when the electric energy is consumed indicates that the estimated target house 1100 indicates the electric energy. The selection representative model 1520 having the total load information similar to the total load information 1080 of the estimation target house 1100 is selected from the two or more representative models 1500 representing the load information of the monitoring house similar to the weather information 1083 at the time of consumption. To be done. Further, the load information 1120 for each device of the estimation target house 1100 is estimated using the selection representative model 1520. As a result, the device-specific load information 1120 is accurately estimated.

It should be noted that, in the present invention, the respective embodiments can be freely combined, or the respective embodiments can be appropriately modified or omitted within the scope of the invention.

Although the present invention has been described in detail, the above description is illustrative in all aspects, and the present invention is not limited thereto. It is understood that innumerable variants not illustrated can be envisaged without departing from the scope of the invention.

1000 load estimation device, 1040 N load information, 1041 N feature amount, 1060 monitoring house, 1100 estimation target house, 1120 device load information, 1160 load information, 1180 total load information, 1181 device load information, 1200 feature quantities, 1220 M_1 primary clusters, 1240 primary clusters, 1260 N(m_1) load information, 1261 N(m_1) feature quantities, 1280 derivation formula, 1300 M_2 secondary clusters, 1320 secondary clusters, 1340 N(m_2) load information, 1360 M_3 tertiary clusters, 1380 tertiary clusters, 1400 N(m_3) load information, 1420 representative models, 1440 M_1*M_2*. M_3 representative models, 1460 selected primary clusters, 1500 2 or more representative models, 1520 selected representative models, 1541 device load ratio calculation unit, 1544 derivation formula creation unit, 1560 N device load ratios, 1580 1 Similarity calculation unit for secondary clustering, 1581 primary clustering unit, 1600 similarity for primary clustering, 1620 primary cluster number, 1660 total load information, 1661 feature amount, 1662 primary cluster number, 1722 secondary clustering Similarity calculation unit, 1723 secondary clustering unit, 1724 similarity calculation unit for tertiary clustering, 1725 tertiary clustering unit, 1726 representative model acquisition unit, 1760 N(m_1) total load information, 1780 secondary clustering Similarity for 1800 N(m_2) total load information, 1840 N(m_2) load information for each device, 1860 similarity for third-order clustering, 1882 representative model creation unit, 1961 primary cluster selection unit, 1963 Representative model selection similarity calculation unit, 1964 Representative model selection unit, 1965 Breakdown calculation unit, 1990 Representative model selection similarity, 2000 load estimation device, 1042 N temperature information, 1082 temperature information, 1043 N Weather information, 1083 Weather information.

Claims (4)

It is each of a plurality of load information, total load information showing the time change of the total load indicating the total amount of electric energy consumed by one consumer, and the electricity consumed by multiple parts of the one consumer. For each load information including partial load information indicating a temporal change of a plurality of partial loads each indicating the magnitude of energy, and each of the load information associated with the characteristic amount indicating the characteristic of the one customer, A partial load ratio calculation unit that calculates a partial load ratio indicating the ratio of partial load,
A first similarity calculator that calculates a first similarity between a plurality of partial load ratios calculated for each of the plurality of load information;
The plurality of load information and the plurality of feature amounts respectively associated with the plurality of load information according to the first similarity are classified into a plurality of primary clusters and belong to each of the plurality of primary clusters. A primary clustering unit that identifies at least one load information and at least one feature amount associated with each of the at least one load information,
Derivation that learns at least one load information and at least one feature amount belonging to each of the plurality of primary clusters, and creates a derivation formula for deriving a primary cluster from the given total load information and the given feature amount. An expression creator,
Calculating a second degree of similarity between a plurality of total load information respectively included in a plurality of load information belonging to each primary cluster which is each of at least one primary cluster included in the plurality of primary clusters; 2 similarity calculation section,
Secondary clustering that classifies a plurality of load information belonging to each primary cluster into a plurality of secondary clusters according to the second similarity and identifies at least one load information belonging to each of the plurality of secondary clusters. Department,
Calculating a third similarity between a plurality of partial load information items respectively included in a plurality of load information items belonging to each secondary cluster item that is each of at least one secondary cluster item included in the plurality of secondary cluster items; 3 similarity calculation section,
At least one load information belonging to each tertiary cluster, which is each of the plurality of tertiary clusters, is classified into a plurality of tertiary clusters according to the third similarity degree. A third-order clustering unit that specifies
A representative model acquisition unit for acquiring a plurality of representative models by creating a representative model representative of at least one load information belonging to each of the tertiary clusters;
The total load information indicating the temporal change in the total load indicating the total amount of electric energy consumed by the estimation target consumer is the given total load information, and the feature amount indicating the characteristic of the estimation target consumer is the A primary cluster selection unit that selects a selected primary cluster derived from the derivation formula when the feature quantity is a given one, from the plurality of primary clusters;
Two or more representative models representing load information belonging to the selected primary cluster are selected from the plurality of representative models, and the total load indicating the total amount of electric energy consumed by the estimation target consumer is changed with time. And a fourth similarity calculation unit for calculating a fourth similarity between the total load information indicating the total load information and the total load information included in each of the two or more representative models,
Based on the fourth degree of similarity, total load information indicating a temporal change in total load indicating the total amount of electric energy consumed by the estimation target consumer, and total load information included in a candidate representative model. A representative model selection unit that selects a selected representative model from the two or more representative models by using a selection criterion in which a candidate representative model is more likely to be selected as the similarity of
Partial load information indicating a temporal change of a plurality of partial loads indicating the magnitude of electric energy consumed by the plurality of parts of the estimation target consumer is reflected so that the partial load information included in the selected representative model is reflected. A calculation part to calculate,
A load estimation device including. Temperature information indicating the temperature when the one consumer consumes electric energy is associated with each load information,
The primary clustering unit classifies a plurality of temperature information items respectively associated with the plurality of load information items into the plurality of primary clusters, and belongs to each of the plurality of primary clusters, the at least one load information item. Specify at least one temperature information associated with
The derivation formula creating unit learns at least one temperature information respectively associated with the at least one load information,
The derivation formula derives the primary cluster from the given total load information, the given feature amount, and the given temperature information,
The selected primary cluster is total load information to which the total load information indicating the total amount of electric energy consumed by the estimation target consumer is given, and the feature amount indicating the feature of the estimation target consumer is It is the given feature quantity and is a primary cluster derived by the derivation formula when the temperature information indicating the temperature when the estimation target consumer consumes electric energy is the given temperature information. The load estimation device according to claim 1. Weather information indicating the weather when the one consumer consumes electric energy is associated with each load information,
The primary clustering unit classifies a plurality of pieces of weather information respectively associated with the plurality of load information into the plurality of primary clusters, and belongs to each of the plurality of primary clusters, the at least one load information. Specify at least one weather information associated with each,
The derivation formula creating unit learns at least one weather information respectively associated with the at least one load information,
The derivation formula derives the primary cluster from the given total load information, the given feature amount, and the given weather information,
The selected primary cluster is total load information to which the total load information indicating the total amount of electric energy consumed by the estimation target consumer is given, and the feature amount indicating the feature of the estimation target consumer is It is the given feature amount, and is a primary cluster derived by the derivation formula when the weather information indicating the weather when the estimation target consumer consumes electric energy is the given weather information. The load estimation device according to claim 1. a) It is each of a plurality of load information, and total load information showing the time change of the total load showing the total amount of electric energy consumed by one consumer, and multiple parts of the one consumer consume For each load information including partial load information indicating a time change of a plurality of partial loads each indicating the magnitude of the electric energy, the characteristic amount indicating the characteristic of the one customer is associated, Calculating a partial load ratio indicating a ratio of a plurality of partial loads;
b) calculating a first similarity between the plurality of partial load ratios respectively calculated for the plurality of load information;
c) According to the first similarity, the plurality of load information and the plurality of feature amounts respectively associated with the plurality of load information are classified into a plurality of primary clusters, and each of the plurality of primary clusters. Belonging to, at least one load information and a step of identifying at least one characteristic amount respectively associated with the at least one load information,
d) Learning at least one load information and at least one feature amount belonging to each of the plurality of primary clusters, and creating a derivation formula for deriving a primary cluster from the given total load information and the given feature amount. The process of
e) calculating a second degree of similarity between a plurality of total load information included in a plurality of load information belonging to each primary cluster, which is each of at least one primary cluster included in the plurality of primary clusters The process of
f) classifying a plurality of load information belonging to each of the primary clusters into a plurality of secondary clusters according to the second similarity, and identifying at least one load information belonging to each of the plurality of secondary clusters. When,
g) calculating a third similarity between a plurality of partial load information items included in a plurality of load information items belonging to each secondary cluster item that is each of at least one secondary cluster item included in the plurality of secondary cluster items. The process of
h) According to the third similarity, a plurality of load information belonging to each of the secondary clusters is classified into a plurality of tertiary clusters, and at least one of the tertiary clusters is one of the tertiary clusters. A step of specifying load information,
i) obtaining a plurality of representative models by creating a representative model representative of at least one load information belonging to each of the tertiary clusters;
j) The total load information indicating the temporal change in the total load indicating the total amount of electric energy consumed by the estimation target consumer is the given total load information, and the feature amount indicating the characteristics of the estimation target consumer. Selecting a selected primary cluster derived from the derivation formula when is the given feature quantity from the plurality of primary clusters,
k) Two or more representative models representing load information belonging to the selected primary cluster are selected from the plurality of representative models, and the total load indicating the total amount of electric energy consumed by the estimation target consumer is calculated. Calculating a fourth similarity between the total load information indicating the time change and the total load information included in each of the two or more representative models;
l) Based on the fourth similarity, total load information indicating the temporal change of the total load indicating the total amount of electric energy consumed by the estimation target consumer and the total load included in the candidate representative model. Selecting a selected representative model from the two or more representative models using a selection criterion in which a candidate representative model is more likely to be selected as the degree of similarity to the information is higher;
m) Partial load indicating a temporal change of a plurality of partial loads indicating the magnitude of electric energy consumed by the plurality of parts of the estimation target consumer so that the partial load information included in the selected representative model is reflected. The process of calculating information,
A load estimation method comprising:

Images