CN115660725A - Method for depicting multi-dimensional energy user portrait - Google Patents
Method for depicting multi-dimensional energy user portrait Download PDFInfo
- Publication number
- CN115660725A CN115660725A CN202210864894.3A CN202210864894A CN115660725A CN 115660725 A CN115660725 A CN 115660725A CN 202210864894 A CN202210864894 A CN 202210864894A CN 115660725 A CN115660725 A CN 115660725A
- Authority
- CN
- China
- Prior art keywords
- information
- target user
- user
- behavior
- portrait
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for depicting a user portrait by a multi-dimensional energy source, which comprises the following steps: acquiring a database in an energy management platform, wherein the database stores basic data of a target user; preprocessing basic data in a database; the method comprises the steps of performing correlation calculation on basic data information of a system user and basic data information of a target user to obtain first portrait information of the target user, performing statistics on user behavior characteristics by using a statistical method of coverage rate and time period preference to determine second portrait information of the target user, and finally performing energy using behavior of the target energy user to form third portrait information of the target user.
Description
Technical Field
The invention relates to the technical field of energy Internet, in particular to a method for depicting a multi-dimensional energy user portrait.
Background
With the continuous deepening of the construction of the energy Internet, energy related industries are actively innovated from top to bottom so as to promote the efficient and reasonable utilization and development of energy. In the construction process of the energy Internet, the coordination and complementary energy among various energy sources is enhanced, and a future comprehensive energy utilization system is gradually evolved into a comprehensive energy system taking users as centers. The importance of the user portrait is increased day by day, and the value of the energy can be exerted to the maximum only by accurately depicting the portrait of the user and clearly analyzing the behavior track of the user.
In the prior art, the portraits of users are mainly characterized in that firstly, users fill in own portraits data, and the problem of the method is that the users are likely to be wrong in filling in or the information filled by the users is not complete; secondly, the user portrait is drawn by using the rules, and the user portrait is drawn by using the geographical position rules of the coming and going every day through the behavior data of the user in the last month; however, the above-mentioned technology for describing the image of the energy user is not mature, which results in inaccurate describing result of the image of the energy user.
Disclosure of Invention
The invention mainly aims to provide a method for depicting a multi-dimensional energy user portrait, and aims to solve the problem that the depicting result of the energy user portrait is not accurate due to the fact that the existing omnibearing information depicting technology for the energy user portrait is not mature.
In order to solve the above problems, the present invention is realized by the following technical scheme:
a method for depicting a user portrait of a multi-dimensional energy source, comprising: and acquiring a database in the energy management platform, wherein the database stores basic data of a target user. And preprocessing the basic data in the database.
And acquiring basic data information of the target user in a first preset time period.
Acquiring basic data information of a system user in a second preset time period; the first preset time period is less than the second preset time period, and the first preset time period is included in the second preset time period.
And performing correlation calculation on the basic data information of the system user and the basic data information of the target user to obtain first image information of the target user.
And calculating second portrait information of the target user by using a statistical method of coverage rate and time period preference.
And predicting to obtain third image information of the target user through the energy consumption behavior information of the target user based on a GRU algorithm model of deep learning.
And determining final portrait information of the target user by combining the first portrait information, the second portrait information and the third portrait information.
Optionally, the step of preprocessing the basic data in the database includes:
and removing missing values and abnormal values in the data information of the target user in the database.
Optionally, the system user is a user who owns the corresponding portrait information. The portrait information includes tag information and corresponding tag content. The basic data information comprises affiliated industry information, household number information, energy consumption information, electric power consumption information, energy supply unit information, power distribution position information, geographical position information and time information. The energy consumption information includes carbon emissions and a corresponding device name.
Optionally, the step of performing correlation calculation on the basic data information of the system user and the basic data information of the target user to obtain the first image information of the target user includes:
associating the affiliated industry information of the target user with the affiliated industry information of the system user, associating the number information of the target user with the number information of the system user, associating the energy consumption information of the target user with the energy consumption information of the system user, associating the power consumption information of the target user with the power consumption information of the system user, associating the energy supply unit information of the target user with the energy supply unit information of the system user, associating the power distribution position information of the target user with the power distribution position information of the system user, associating the geographic position information of the target user with the geographic position information of the system user, and associating the time information of the target user with the time information of the system user to obtain an association comparison result;
and determining first image information of the target user based on the correlation comparison result.
Optionally, the step of determining the first image information of the target user based on the correlation comparison result includes: and based on the correlation comparison result, marking the system user with the distance to the geographic position of the target user smaller than a preset distance as a related user. And acquiring label content of the target user consistent with the related user based on the portrait information of the related user, and marking the label content as the related content. And setting the related content and the corresponding label information as the first image information of the target user.
Optionally, the step of calculating the second portrait information of the target user by using a statistical method of coverage and time period preference includes: and defining a certain type of business behavior and behavior state of the target user at a certain moment as a behavior tag T, wherein the behavior tag T is the second portrait information. The behavior characteristics represented by each behavior tag T are represented by frequency, average, coverage, deviation, average time interval, period characteristics and period preference characteristics. The image information includes behavior tag information and corresponding tag content. The business behaviors with the behavior tag T include the following categories: the monthly energy consumption, annual energy consumption maximum, annual energy consumption minimum, energy consumption variable quantity, energy consumption change rate, energy consumption peak period and payment condition of the target user.
Optionally, the step of calculating the second image information of the target user by using the statistical method of the coverage rate and the time period preference further includes:
the target user u generates the business behavior at of the jth behavior label in a certain time period j The coverage rate is represented by createdratio, and the coverage rate createdratio is calculated by the formula:
wherein ET-ST is the statistical time length, sum (at) j ,u) ET-ST Indicating that the target user u takes the business action at in the ET-ST j The sum of the times of (c); j represents all of the selected target users uThe jth behavior tag, at, inside the behavior tags j Representing the business behavior of the jth behavior label of the target user u; sum (at) i ,u) ET-ST The service action at occurs in the period of time indicated as ET-ST for the target user u i The sum of the times of (a), i is in the range of [0,n]I represents the selection of the ith behavior tag, at, from all the behavior tags of the target user u i Representing the business behavior of the ith behavior tag of the target user u; the target user u generates the business behavior at of the jth behavior label j The average occurrence time interval is represented by average (d), and the average (d) is calculated by the following formula:
where n denotes the occurrence of at by the target user u j The number of business actions; d is a radical of J Indicating that target user u occurred at j J-th time interval in business behavior; the target user u generates the service behavior at of the jth behavior label j Degree of deviation of
The calculation formula is as follows:
where n denotes the occurrence of at by the target user u j The number of business actions; d J Indicating that target user u occurred at j Jth time interval in business behavior; the target user u is subjected to the business behavior at of the jth behavior label j Is divided into n line intervals D 1， D 2， ...D n ，
Said traffic behavior at j Periodic period (at) of (2) j U) is calculated as
Wherein, sum (d) I U) represents the behavior interval d of the target user u I The number of occurrences, I, is in the range of [1,n]; d I Indicating that target user u occurred at j The I time interval of business behavior; if the business behavior at j Without periodicity, it is denoted by 0. If the traffic behavior at of the target user u is j Number of times sum (d) occurred in the period I U) accounts for 60% of the total number of occurrences of the behavior, then the business behavior at j Has a time period preference, and the time period preference characteristic TF is the time period.
Optionally, the step of predicting the predicted tag information by the GRU algorithm based on deep learning to obtain third image information of the target user includes: and marking the label information to be predicted of the target user as a label to be predicted. The label content to be predicted comprises: predicting the energy consumption, the energy consumption behavior and the energy consumption change of a target user at a certain future time; establishing a GRU algorithm model based on deep learning; taking the labels to be predicted of the target users and the corresponding label contents as training data to train the GRU algorithm model based on deep learning; and predicting to-be-pre-labeled labels and corresponding label contents of the target users through the trained GRU algorithm model based on deep learning, and using the labels and the corresponding label contents as the third image information.
Optionally, the step of training the GRU algorithm model based on deep learning by using the information to be predicted of the target user as training data includes: state h transmitted by the last node t-1 And the input value x of the current node t t To obtain two gating states. After obtaining the gating state, reset gating is used to obtain data h after reset t-1’ Then the data h is processed t-1’ And the input value x t And (6) splicing.
Scaling the spliced data to [ -1~1 ] by tanh activation function]To obtain the candidate state h of the current node state ’ . Candidate state h of current node state ’ Simultaneously, forgetting and memorizing.
Optionally, the gating state includes an update gate state and a reset gate state; the updating gate is used for controlling the degree of the state information of the previous moment retained in the current state, and the larger the value of the updating gate is, the more the state information of the previous moment is retained; the reset gate is used to determine whether to combine the current state with previous information, with smaller values of the reset gate indicating more information to ignore.
The current node state h t The state updating method comprises the following steps:
wherein z is the update gate state,
, h ’ representing a candidate state at the current time; x is a radical of a fluorine atom t An input value representing the current node t;Wthe parameters representing the memory gated neurons are learned during the training process.
Optionally, the step of determining final portrait information of the target user by combining the first portrait information, the second portrait information and the third portrait information includes: and packaging the first image information, the second image information and the third image information to form the final image information corresponding to the target user.
The invention has at least one of the following advantages:
according to the multi-dimensional energy user portrait depicting method, the first portrait information of a target user is determined through the relevant data information dimensionality, the second portrait information of the target user is determined through a statistical method of coverage rate and time period preference, finally, the label information needing to be predicted of the target energy user is obtained through deep learning, and further the third portrait information of the target user is formed. The basic data information adopted by the invention belongs to the static energy utilization information of the target user. Now, information such as energy suppliers is added, which is not only convenient for the target users to use, but also provides information basis for the suppliers to provide future energy utilization service (i.e. third image) of the target users.
Drawings
FIG. 1 is a flowchart illustrating a method for depicting a user portrait of a multi-dimensional energy source according to a first embodiment of the present invention;
FIG. 2 is a diagram of input/output structures of GRU algorithm of a sixth embodiment of a method for depicting a multi-dimensional energy user portrait according to the present invention;
FIG. 3 is a schematic structural diagram of a reset gate and an update gate according to a seventh embodiment of the method for depicting a multi-dimensional energy user image according to the present invention;
FIG. 4 is a diagram of a method for depicting a multi-dimensional energy user image according to a seventh embodiment of the present invention t Schematic process diagram of (1).
Detailed Description
The following describes a method for depicting a user portrait of a multi-dimensional energy source according to the present invention in detail with reference to the accompanying drawings and the following detailed description. The advantages and features of the present invention will become more apparent from the following description. It is to be noted that the drawings are in a very simplified form and are all used in a non-precise scale for the purpose of facilitating and distinctly aiding in the description of the embodiments of the present invention. To make the objects, features and advantages of the present invention comprehensible, reference is made to the accompanying drawings. It should be understood that the structures, ratios, sizes, and the like shown in the drawings and described in the specification are only used for matching with the disclosure of the specification, so as to be understood and read by those skilled in the art, and are not used to limit the implementation conditions of the present invention, so that the present invention has no technical significance, and any structural modification, ratio relationship change or size adjustment should still fall within the scope of the present invention without affecting the efficacy and the achievable purpose of the present invention.
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the structures shown in the drawings without creative efforts.
Example one
As shown in fig. 1, the present embodiment provides a method for depicting a user portrait of a multi-dimensional energy source, including:
step S100: and acquiring a database in the energy management platform, wherein the database stores basic data of a target user. The energy management platform can also be called as an energy brain, and is a smart city energy cloud platform created by applying technologies such as big data analysis, artificial intelligence and cloud computing.
Step S101, preprocessing the basic data in the database, in this embodiment, the step S101 specifically includes: and removing missing values and abnormal values in the data information of the target user in the database.
It is understood that the preprocessed data is tag-definable and operable. In this embodiment, the basic data and the energy (i.e., static and dynamic data) are included, the static data is used as the basis for the correlation calculation of the first image information, and the dynamic data is used as the basis for the correlation calculation of the second image information and the third image information, and the statistics, calculation and prediction are performed.
And step S102, acquiring basic data information of the target user in a first preset time period.
Acquiring basic data information of a system user in a second preset time period; the first preset time period is less than the second preset time period, and the first preset time period is included in the second preset time period. And performing correlation calculation on the basic data information of the system user and the basic data information of the target user to obtain first image information of the target user.
In the present embodiment, the first preset time period is preferably 1 month, but the present invention is not limited thereto. The basic data information can be obtained remotely through a smart meter.
For example, when the target user a consumes energy, the energy consumption information is acquired as { "emission factor": "3", "energy _ name": "energy of a small famous family", and electric power consumption information is { "electric quantity grade": "first class", "consumption grade": "high", geographical location information is { "wm629s9": "zone a No. 1000"; the traffic place name: path a "}, the first preset time is: 6 months 1 day-7 months 1 day in 2021.
The system user is a user with corresponding portrait information, and the portrait information comprises tag information and corresponding tag content; the basic data information comprises affiliated industry information, household number information, energy consumption information, electric power consumption information, energy supply unit information, power distribution position information, geographical position information and time information. The first preset time period is less than the second preset time period, the first preset time period is included in the second preset time period, and the energy consumption information comprises carbon emission and a corresponding equipment name.
In the present embodiment, the second predetermined period of time is 6 months, but the present invention is not limited thereto.
For example, the energy consumption information of a certain system user Y is { "emission factor": 3, energy _ name, energy of the system Y0 family, and electric power consumption information as { "electric quantity grade": "first class", "consumption grade": "low" }, geographical location information of "wm629s9": "zone a 10010"; a name of a traffic place; path a "}, the second preset time is: 1/month 1/7/month 1/2021.
For example: the tag information in the portrait information includes: basic data information, turnover, industry category, city classification, power supply voltage, load property, energy utilization category, energy activity, energy consumption level, carbon emission amount, energy proportion, electric quantity level, seasonal power consumption peak, power supply quality perception, illegal power utilization degree, power utilization category, energy consumption type, meter reading mode, meter reading period, bill issuing day, payment deadline day and consumption level.
Each piece of label information corresponds to label content, for example, the label content of the turnover is XX element; the tag content of the supply voltage is 220V, etc.
In this embodiment, the first image information is image information of a system user who is close to the target user, and is consistent with some image information by the user who is close to the target user, for example, seasonal power consumption peak, power level, power supply voltage, city classification, and the like in the basic data information.
Step S103: and calculating second portrait information of the target user by utilizing a statistical method of coverage rate and time period preference.
The coverage rate represents the proportion of the occurrence times of the business behaviors in a certain time period to the sum of the occurrence times of the same business behaviors. The behavior occurrence average time interval is the average of the behavior tag occurrence time intervals. The deviation degree, i.e., the standard deviation of the time interval of occurrence of the behavior tag, represents the time uniformity of a certain behavior generated by the user, and the lower the deviation degree, the behavior may be a periodic behavior. Periodicity is used to measure whether a user's behavior is periodic. The period preference feature represents a period preference resulting from user behavior. The above features are described from a history state and a near state, and the time characteristics of a certain behavior are highlighted.
In this embodiment, the step S103 includes: the method comprises the following steps of defining certain service behaviors and behavior states of a target user at a certain moment as behavior tags T, namely using the behavior tags T as second portrait information, representing behavior characteristics represented by each behavior tag T by frequency, average value, coverage rate, deviation, average time interval, periodic characteristics and time period preference characteristics, wherein the service behaviors with the behavior tags T comprise the following classes: the monthly energy consumption, annual energy consumption maximum, annual energy consumption minimum, energy consumption variable quantity, energy consumption change rate, energy consumption peak period and payment condition of the target user.
In this embodiment, the step S103 further includes: the target user u generates the business behavior at of the jth behavior label in a certain time period j Coverage ofExpressed by createdratio, the coverage createdratio is calculated by the formula:
wherein ET-ST is the statistical time length, sum (at) j ,u) ET-ST Indicating that the target user u takes the business action at in the ET-ST j The sum of the times of (c); j represents the j th behavior label, at, among all the behavior labels of the selected target user u j Representing the business behavior of the jth behavior label of the target user u; sum (at) i ,u) ET-ST The service action at occurs in the period of time indicated as ET-ST for the target user u i The value range of i is [0,n ]]I represents the selection of the ith behavior tag, at, from all the behavior tags of the target user u i And (3) representing the business behavior of the ith behavior tag of the target user u.
The target user generates the business behavior at of the jth behavior label j The average occurrence time interval is represented by average (d), and the average (d) is calculated by the following formula:
where n denotes the occurrence of at by the target user u j The number of business actions; d is a radical of J Indicating that target user u occurred at j J-th time interval in business behavior;
the target user u generates the service behavior at of the jth behavior label j Degree of deviation of
The calculation formula is as follows:
where n denotes the occurrence of at by the target user u j The number of business actions; d J Indicating that target user u occurred at j Jth time interval in business behavior;
the target user u is subjected to the business behavior at of the jth behavior label j Is divided into n line intervals D 1， D 2， ...D n 。
Said traffic behavior at j Periodic period (at) of (2) j U) is calculated as
In the formula (d) I U) represents the behavior interval d of the target user u I The number of occurrences, I, is in the range of [1,n]; d I Indicating that target user u occurred at j The ith time interval when the business acts. If the business behavior at j Without periodicity, it is denoted by 0.
If the traffic behavior at of the target user u is at j Number of times sum (d) occurred in the period I U) accounts for 60% of the total number of occurrences of the behavior, then the business behavior at j Has a time period preference, and the time period preference characteristic TF is the time period. Step S104: and predicting to obtain third image information of the target user through partial image information of the system user by the GRU algorithm model based on deep learning.
Specifically, the portrait information of the target user is obtained through correlation calculation and a statistical method of coverage rate and time period preference, and whether partial label information is not available or not is not available, so that partial labels of system users with existing portrait information can be deeply learned. Firstly, a GRU algorithm model based on deep learning is generated, then labels to be predicted of target users and corresponding label contents are used as training data to train the GRU algorithm model based on deep learning, wherein the label contents to be predicted mainly include energy consumption prediction, energy consumption behavior prediction and energy consumption change prediction at a certain time in the future.
Step S105: the first, second, and third portrait information are combined to determine final portrait information for the target user.
Through the steps, the first image information, the second image information and the third image information of the target user can be obtained, and further the final image information of the target user can be obtained.
According to the method for depicting the multi-dimensional energy user portrait, the first portrait information of the target user is determined through the relevant data information dimension, the second portrait information of the target user is calculated and determined through a statistical method of coverage rate and time period preference, finally the label information needing to be predicted of the target energy user is obtained through deep learning, and the third portrait information of the target user is formed.
Example two
Based on the first embodiment, the present embodiment is different from the first embodiment in that the step S102 further includes the following steps:
step S1021: associating the affiliated industry information of the target user with the affiliated industry information of the system user, associating the number information of the target user with the number information of the system user, associating the energy consumption information of the target user with the energy consumption information of the system user, associating the power consumption information of the target user with the power consumption information of the system user, associating the energy supply unit information of the target user with the energy supply unit information of the system user, associating the power distribution position information of the target user with the power distribution position information of the system user, associating the geographic position information of the target user with the geographic position information of the system user, and associating the time information of the target user with the time information of the system user to obtain an association comparison result.
Step S1022: and determining first image information of the target user based on the association result.
Specifically, the correlation result is to compare the energy consumption information of the target user with the industry information of the system user, compare the energy consumption information of the target user with the energy consumption information of the system user, compare the energy consumption information of the target user with the energy supply unit information of the system user, compare the energy consumption information of the target user with the power distribution position information of the system user, compare the power consumption information of the target user with the power consumption information of the system user, compare the geographical position information of the target user with the geographical position information of the system user, and compare the time information of the target user with the time information of the system user.
EXAMPLE III
Based on the second embodiment, the method for depicting a user portrait of a multi-dimensional energy source proposed in this embodiment is different from the second embodiment in that the step S1022 includes the following steps:
step S10221: and based on the correlation result, marking the system user with the distance to the geographic position of the target user smaller than a preset distance as a related user.
Specifically, the reasonable preset distance may be 1 km, but the present invention is not limited thereto, that is, the system user whose distance from the geographic position of the target user is less than 1 km is marked as the relevant user by comparing the geographic position information of the target user with the geographic position information of the system user.
And if a plurality of system users with the distance less than the preset distance exist, marking the system user with the minimum distance as a related user. Or to mark system users for which the energy consumption information and the power consumption information are closer as related users.
Step S10222: and acquiring label content of the target user consistent with the related user based on the portrait information of the related user, and marking the label content as the related content.
Specifically, because the target user and the related user are relatively close to each other, some tag contents are consistent, for example: seasonal power consumption peak, electric quantity grade, power supply voltage and city classification; this marks these tagged contents as relevant contents.
Step S10223: and setting the related content and the corresponding label information as first image information of a target user.
Specifically, the seasonal power consumption peak, the power level, the power supply voltage and the city classification of the related users are used as the first image information of the target users.
Example four
In a fourth embodiment, based on the first embodiment, the step S104 of the method for depicting a portrait of a multi-dimensional energy user further includes the following steps:
step S1041: and marking label information which is not owned by the target user and owned by the related user as a label to be predicted.
Specifically, for example: the portrait information of the target user does not have label information of energy consumption level, and the portrait information of the related user is used for the label information, so that the energy consumption level is the label to be predicted corresponding to the target user.
Step S1042: and establishing a GRU algorithm model based on deep learning.
Step S1043: and taking the labels to be predicted of the related users and the corresponding label contents as training data to train the GRU algorithm model based on deep learning.
Step S1044: and predicting to-be-pre-labeled labels and corresponding label contents of the target users through the trained GRU algorithm model based on deep learning, and using the to-be-pre-labeled labels and the corresponding label contents as the third image information.
Specifically, the first portrait information and the second portrait information of the target user obtained through the associated data information and the nearest neighbor algorithm dimension may or may not reflect the portrait of the target user comprehensively, namely whether the target user has partial label information or not, and partial labels of related users with existing portrait information can be learned through depth. Firstly, model training is carried out on part of label information of related users to generate a training module, and part of labels are labels which are not carried by target users. In this patent, the system has predicted most of the portrait information of the target user through multiple dimensions, and the rest can be predicted through deep learning.
For example, through the steps, the system predicts the label information of the target user such as industry classification, electric quantity grade, business turnover, city classification and the like, but the label information of the energy activity does not exist, and can perform deep learning on the label information of the energy activity of the system user related to the target user and the corresponding label content to obtain the label information of the energy activity of the target user and the label content. And using the label information of the energy activity and the paired label contents as third image information of the target user.
The GRU algorithm (Gate recovery Unit) is one of Recurrent Neural Networks (RNN). Like LSTM (Long-Short Term Memory), it is proposed to solve the problems of Long-Term Memory and gradient in back-propagation.
RNNs are suitable for analyzing time series data because RNNs introduce a cyclic cell structure in the network and allow hiding internal connections between cells, making it possible to explore the temporal relationship between non-contiguous data. However, RNNs suffer from the problem of gradient disappearance, resulting in RNNs losing the ability to learn information that was long in the past as time intervals increase.
The proposed LSTM neural network solves the problem of RNN gradient disappearance, has been widely applied in the field of predicting time series data, and many variants have evolved in recent years according to different requirements. GRU as a variation of LSTM, adopts gated recurrent neural network structure, has less training parameters compared with LSTM, and maintains the prediction effect of LSTM.
The GRU algorithm internal unit is very similar to the internal unit of the LSTM, except that the GRU combines the input gate and the forgetting gate in the LSTM into a single update gate. Thus, there are only two gate structures in the GRU, the refresh gate and the reset gate respectively. The update gate is used for controlling the degree of the state information of the previous moment retained in the current state, and the larger the value of the update gate is, the more the state information of the previous moment is retained. The reset gate is used to determine whether to combine the current state with previous information, with smaller values of the reset gate indicating more information to ignore.
As shown in fig. 2, the input-output structure of the GRU algorithm is the same as that of a normal RNN.
With an input value x for the current node t t And hidden state h passed down by the previous node t-1 This hidden state contains information about the previous node.
Combining input values x t And hidden state h t-1 The GRU algorithm will obtain the output y of the current hidden node t And a hidden state h passed to the next node t 。
State h transmitted by the last node t-1 And the input value x of the current node t t To obtain two gating states.
Specifically, as shown in fig. 3, r controls the gating of reset (reset gate), and z controls the gating of update (update gate).
After obtaining the gating state, reset gating is used to obtain data h after reset t-1’ Then the data h t-1’ And the input value x t And (6) splicing. It can thus be understood that state h t-1 After reset gating processing, the state is changed into a value h t-1’ 。
Scaling the spliced data to [ -1~1 ] by tanh activation function]To obtain the candidate state h of the current node state ’ 。
Specifically, the current node state h is obtained t The process of (a) is shown in fig. 4. H herein t Mainly containing x of the current input value t And (4) data. In a specific way for h t Adding to the current hidden state corresponds to "memorizing the state at the current time".
Candidate state h of current node state ’ Simultaneously, forgetting and memorizing.
Specifically, this step uses the previously obtained update gate z (update gate), which is called the "update memory phase".
EXAMPLE five
In a fifth embodiment of the present invention, based on the fourth embodiment, the gating state in the method for depicting a multi-dimensional energy user portrait provided in this embodiment includes an update gate state and a reset gate state; the updating gate is used for controlling the degree of the state information of the previous moment retained in the current state, and the larger the value of the updating gate is, the more the state information of the previous moment is retained; the reset gate is used to determine whether to combine the current state with previous information, with smaller values of the reset gate indicating more information to ignore.
EXAMPLE six
Based on five embodiments, in the method for depicting a multi-dimensional energy user portrait provided by this embodiment, the current node state h is represented t The state updating method comprises the following steps:
wherein z is the update gate state,
, h ’ representing a candidate state at the current time; x is the number of t An input value representing the current node t;Wparameters representing the memory gate neurons are learned during the training process; h is ’ Containing mainly x as current input t The data is equivalent to memorizing the state at the current time.
EXAMPLE seven
Based on the first embodiment or the sixth embodiment, the step S105 of the method for depicting a user portrait by using multi-dimensional energy source of the present embodiment includes the following steps:
and packaging the first portrait information, the second portrait information and the third portrait information to form final portrait information corresponding to the target user.
Specifically, the portrait information of the target user is obtained through multi-dimensional calculation, similar data information may exist, specifically, the system extracts the similar data information, comprehensively sorts the data information, and counts the most accurate data information.
Then, the tag information similar to the first image information, the second image information and the third image information is screened for integration; the first, second, and third image information are combined to determine final image information for the target user.
In summary, the embodiment utilizes the rich and reliable data source of the "energy brain" to not only reduce the situation that the user portrait information structure of the energy target is inaccurate, but also depict more accurate and complete user portrait information. The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
Through the description of the foregoing embodiments, it is clear to those skilled in the art that the method of the foregoing embodiments may be implemented by software plus a necessary general hardware platform, and certainly may also be implemented by hardware, but in many cases, the former is a better implementation. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, wherein the software product is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk), and includes instructions for enabling a terminal (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present invention.
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.
While the present invention has been described in detail with reference to the preferred embodiments, it should be understood that the above description should not be taken as limiting the invention. Various modifications and alterations to this invention will become apparent to those skilled in the art upon reading the foregoing description. Accordingly, the scope of the invention should be determined from the following claims.
Claims (10)
1. A method for depicting a user portrait of a multi-dimensional energy source, comprising:
acquiring a database in an energy management platform, wherein the database stores basic data of a target user;
preprocessing the basic data in the database;
acquiring basic data information of a target user in a first preset time period;
acquiring basic data information of a system user in a second preset time period; the first preset time period is less than the second preset time period, and the first preset time period is included in the second preset time period;
performing correlation calculation on the basic data information of the system user and the basic data information of the target user to obtain first image information of the target user;
calculating second image information of the target user by using a statistical method of coverage rate and time period preference;
predicting to obtain third image information of the target user through the energy consumption behavior information of the target user based on a GRU algorithm model for deep learning;
and determining final portrait information of the target user by combining the first portrait information, the second portrait information and the third portrait information.
2. A method of multi-dimensional energy user representation portrayal as claimed in claim 1, wherein said step of pre-processing said underlying data in said database comprises:
and removing missing values and abnormal values in the data information of the target user in the database.
3. The method of claim 1, wherein the user representation of the multi-dimensional energy source is generated by a computer,
the system user is a user with corresponding portrait information;
the portrait information includes tag information and corresponding tag content;
the basic data information comprises affiliated industry information, household number information, energy consumption information, electric power consumption information, energy supply unit information, power distribution position information, geographical position information and time information;
the energy consumption information includes carbon emissions and a corresponding device name.
4. The method of claim 3, wherein the user representation of the multi-dimensional energy source is generated by a computer,
the step of performing correlation calculation on the basic data information of the system user and the basic data information of the target user to obtain the first image information of the target user comprises the following steps:
associating the affiliated industry information of the target user with the affiliated industry information of the system user, associating the number information of the target user with the number information of the system user, associating the energy consumption information of the target user with the energy consumption information of the system user, associating the power consumption information of the target user with the power consumption information of the system user, associating the energy supply unit information of the target user with the energy supply unit information of the system user, associating the power distribution position information of the target user with the power distribution position information of the system user, associating the geographic position information of the target user with the geographic position information of the system user, and associating the time information of the target user with the time information of the system user to obtain an association comparison result;
and determining first image information of the target user based on the correlation comparison result.
5. The method of claim 4, wherein the user representation of the multi-dimensional energy source is generated by a computer,
the step of determining the first image information of the target user based on the correlation comparison result comprises:
based on the correlation comparison result, marking the system user with the distance to the geographic position of the target user smaller than a preset distance as a related user;
acquiring label content of the target user consistent with the related user based on the portrait information of the related user, and marking the label content as related content;
and setting the related content and the corresponding label information as the first image information of the target user.
6. The method of claim 5, wherein the user representation of the multi-dimensional energy source is generated by a computer,
the step of calculating the second portrait information of the target user using the statistical method of the coverage rate and the time period preference includes:
defining a certain type of service behaviors and behavior states of the target user at a certain moment as a behavior tag T, wherein the behavior tag T is the second portrait information;
representing the behavior characteristics represented by each behavior label T by frequency, average value, coverage rate, deviation degree, average time interval, periodic characteristics and period preference characteristics;
the business behaviors with the behavior tag T include the following categories: the monthly energy consumption, annual energy consumption maximum, annual energy consumption minimum, energy consumption variable quantity, energy consumption change rate, energy consumption peak period and payment condition of the target user.
7. The method for multi-dimensional energy user representation characterization according to claim 6 wherein the step of calculating the second representation information of the target user using statistical methods of coverage and time period preference further comprises:
the target user u generates the business behavior at of the jth behavior label in a certain time period j The coverage rate is represented by createdratio, and the coverage rate createdratio is calculated by the formula:
wherein ET-ST is the statistical time length, sum (at) j ,u) ET-ST Indicating that the target user u takes the business action at in the ET-ST j The sum of the times of (c); j represents the j th behavior label, at, among all the behavior labels of the selected target user u j Representing the business behavior of the jth behavior label of the target user u;
sum(at i ,u) ET-ST the service action at occurs in the period of time indicated as ET-ST for the target user u i The sum of the times of (a), i is in the range of [0,n]I represents the selection of the ith behavior tag, at, from all the behavior tags of the target user u i Representing the business behavior of the ith behavior tag of the target user u;
the target user u generates the business behavior at of the jth behavior label j The appearance average time interval is represented by average (d), and the calculation formula of average (d) is as follows:
where n denotes the occurrence of at by the target user u j The number of business actions; d J Indicating that target user u occurred at j J-th time interval in business behavior;
the target user u generates the service behavior at of the jth behavior label j Degree of deviation of
The calculation formula is as follows:
where n denotes the occurrence of at by the target user u j The number of business actions; d J Indicating that target user u occurred at j Jth time interval in business behavior;
the target user u is subjected to the business behavior at of the jth behavior label j Is divided into n line intervals D 1 ，D 2 ，...D n Then said traffic behavior at j Periodic period (at) of (2) j U) is calculated as
Wherein, sum (d) I U) represents the behavior interval d of the target user u I The number of occurrences, I, is in the range of [1,n]；d I Indicating that target user u occurred at j The I time interval of business behavior;
if the business behavior at j Without periodicity, it is represented by 0;
if the traffic behavior at of the target user u is at j Number of times sum (d) occurred in the period I U) accounts for 60% of the total number of occurrences of the behavior, then the business behavior at j Has a time period preference, and the time period preference characteristic TF is the time period.
8. The method of claim 7, wherein the user representation of the multi-dimensional energy source is generated by a computer,
the step of predicting the predicted label information by the GRU algorithm based on deep learning to obtain the third image information of the target user comprises the following steps:
marking the label information to be predicted of the target user as a label to be predicted;
the label content to be predicted comprises: predicting the energy consumption, the energy consumption behavior and the energy consumption change of a target user at a certain future time;
establishing a GRU algorithm model based on deep learning;
taking the labels to be predicted of the target users and the corresponding label contents as training data to train the GRU algorithm model based on deep learning;
and predicting to-be-pre-labeled labels and corresponding label contents of the target users through the trained GRU algorithm model based on deep learning, and using the labels and the corresponding label contents as the third image information.
9. The method of claim 8, wherein the user representation is generated by a user,
the step of training the GRU algorithm model based on deep learning by taking the information to be predicted of the target user as training data comprises the following steps:
state h transmitted by the last node t-1 And the input value x of the current node t t To obtain two gating states;
after obtaining the gating state, reset gating is used to obtain data h after reset t-1’ Then the data h is processed t-1’ And the input value x t Splicing is carried out;
scaling the spliced data to the range of [ -1 ] through a tanh activation function to obtain a current node state candidate state h';
and simultaneously forgetting and memorizing the candidate state h' of the current node state.
10. The method of claim 9, wherein the user profile is a multi-dimensional representation of a user of the energy source,
the gating state comprises an update gate state and a reset gate state; the updating gate is used for controlling the degree of the state information of the previous moment retained in the current state, and the larger the value of the updating gate is, the more the state information of the previous moment is retained; the reset gate is used for determining whether to combine the current state with the previous information, and the smaller the value of the reset gate is, the more the ignored information is;
the current node state h t The state updating method comprises the following steps:
h t ＝z⊙h t-1 +(1-z)⊙h’
wherein z is the update gate state,
h' represents a candidate state at the current moment; x is the number of t An input value representing the current node t; w represents the parameters of the memory gate neuron, which are obtained by learning in the training process;
the step of determining the final portrait information of the target user by combining the first portrait information, the second portrait information and the third portrait information comprises:
and packaging the first image information, the second image information and the third image information to form the final image information corresponding to the target user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210864894.3A CN115660725A (en) | 2022-07-21 | 2022-07-21 | Method for depicting multi-dimensional energy user portrait |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210864894.3A CN115660725A (en) | 2022-07-21 | 2022-07-21 | Method for depicting multi-dimensional energy user portrait |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115660725A true CN115660725A (en) | 2023-01-31 |
Family
ID=85023770
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210864894.3A Pending CN115660725A (en) | 2022-07-21 | 2022-07-21 | Method for depicting multi-dimensional energy user portrait |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115660725A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112465565A (en) * | 2020-12-11 | 2021-03-09 | 加和（北京）信息科技有限公司 | User portrait prediction method and device based on machine learning |
| CN115955452A (en) * | 2023-03-15 | 2023-04-11 | 上海帜讯信息技术股份有限公司 | 5G message pushing method and device based on multi-turn conversation intention recognition |
| CN118397193A (en) * | 2024-06-24 | 2024-07-26 | 深圳供电局有限公司 | Urban electricity image display method, apparatus, electronic device and readable medium |
-
2022
- 2022-07-21 CN CN202210864894.3A patent/CN115660725A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112465565A (en) * | 2020-12-11 | 2021-03-09 | 加和（北京）信息科技有限公司 | User portrait prediction method and device based on machine learning |
| CN112465565B (en) * | 2020-12-11 | 2023-09-26 | 加和（北京）信息科技有限公司 | User portrait prediction method and device based on machine learning |
| CN115955452A (en) * | 2023-03-15 | 2023-04-11 | 上海帜讯信息技术股份有限公司 | 5G message pushing method and device based on multi-turn conversation intention recognition |
| CN115955452B (en) * | 2023-03-15 | 2023-06-06 | 上海帜讯信息技术股份有限公司 | 5G message pushing method and device based on multi-round conversation intention recognition |
| CN118397193A (en) * | 2024-06-24 | 2024-07-26 | 深圳供电局有限公司 | Urban electricity image display method, apparatus, electronic device and readable medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| Lin et al. | Short-term load forecasting based on LSTM networks considering attention mechanism | |
| CN115660725A (en) | Method for depicting multi-dimensional energy user portrait | |
| US20170308934A1 (en) | Management method of power engineering cost | |
| Yilmaz et al. | Synthetic demand data generation for individual electricity consumers: Generative Adversarial Networks (GANs) | |
| CN112967088A (en) | Marketing activity prediction model structure and prediction method based on knowledge distillation | |
| CN102780765A (en) | Cloud manufacturing service resource match and combination method based on performance fusion | |
| CN112381673B (en) | Park electricity utilization information analysis method and device based on digital twin | |
| CN115145993A (en) | Railway freight big data visualization display platform based on self-learning rule operation | |
| CN111489259A (en) | Stock market risk prediction intelligent implementation method based on deep learning | |
| CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
| Yang et al. | An ensemble prediction system based on artificial neural networks and deep learning methods for deterministic and probabilistic carbon price forecasting | |
| CN112330052A (en) | Distribution transformer load prediction method | |
| CN112508734B (en) | Method and device for predicting power generation capacity of power enterprise based on convolutional neural network | |
| CN115600729A (en) | Grid load prediction method considering multiple attributes | |
| CN115983485A (en) | Prediction method and system for carbon emission of power plant | |
| CN109657846A (en) | Power grid alternative subsidy scale impact factor screening technique | |
| CN113762591B (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning | |
| CN114004513A (en) | Demand prediction method, system and storage medium | |
| Li | Prediction of house price index based on machine learning methods | |
| Lin | [Retracted] Big Data Technology in the Macrodecision‐Making Model of Regional Industrial Economic Information Applied Research | |
| Zhao et al. | A global forecasting method of heterogeneous household short-term load based on pre-trained autoencoder and deep-LSTM model | |
| Alikhani et al. | Optimal demand response programs selection using CNN‐LSTM algorithm with big data analysis of load curves | |
| Han et al. | Adaptive individual residential load forecasting based on deep learning and dynamic mirror descent | |
| Chen et al. | Multi-factor stock selecting model based on residual net and LSTM deep learning algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |