JPH0886490A - Predicting equipment of thermal load - Google Patents

Abstract

PURPOSE: To enable highly accurate prediction of a thermal load of the next day without relying on manual efforts, by formulating the nonlinear relationship between factors of change of a thermal load such as week days and weather and the thermal load. CONSTITUTION: While a thermal load consumed by a heat consuming apparatus 5 is detected, a weather measured value and a weather predicted value of an atmospheric temperature etc., are inputted and these are stored by a data storage means 11. A calender data generating means 12 generates week day data, while a prediction model learning means 13 learns weighting factors of a day thermal load prediction model and a time thermal load pattern prediction model by using selectively the weather measured value, the weather predicted value, the thermal load, the week day data and thermal load feature amount data on the morning and afternoon obtained from measured value of a time thermal load, which are stored, and stores them. Based on the weighting factors and the prediction models being stored, a time thermal load pattern is predicted and object heat source machines 3 are controlled by using this predicted time thermal load pattern.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a heat load predicting device for predicting a heat load to be consumed the next day in a heat consuming device such as an air conditioner installed in various buildings including a building.

[0002]

2. Description of the Related Art In buildings such as buildings, houses and other public facilities, an air conditioner having a heat storage tank is installed.
In the heat storage system using electricity, part or all of the heat load consumed in the daytime is stored as cold water or hot water at night, and the heat load of the heat source device can be reduced in the daytime, so that the installed capacity of the heat source device can be reduced. .

Further, in the case of heating and cooling by electricity, by making a heat storage adjustment contract for business use with an electric power company, it is possible to save running costs by utilizing inexpensive nighttime electric power.

Therefore, in such a cooling and heating apparatus, from the viewpoint of efficiently using the heat storage tank, the heat load is predicted to be consumed on the current day (prediction day) before the start of the business heat storage adjustment contract on the previous day. Then, during the contract time period (night), the amount of heat is stored without excess or deficiency based on the prediction. Then, on the day, the heat source device is controlled according to the prediction, and cold water, hot water and steam are supplied to the cooling and heating device which is a heat consuming device.

On the other hand, in the case of heating and cooling by gas, it takes time to preheat the boiler and the heat source equipment, so from the meaning of stable heat supply and efficient heat production, the heat load consumed on the day is predicted. However, it will be necessary to prepare for operation so that it can respond sufficiently.

Therefore, conventionally, as a means for predicting the heat load of the current day, which is the predicted date, a reference heat load for each month is set in advance, and this reference heat load is linearly corrected based on the actual heat load and the outside air temperature. It has been attempted to predict the heat load by performing the above, or to predict the heat load for cooling and heating by summing the heat amounts obtained for each factor such as the amount of solar radiation, the amount of heat entering the outside air, the amount of heat generated indoors, and the amount of heat stored indoors. .

[0007]

By the way, in the heat load prediction method as described above, it is necessary to set constants suitable for the heat source equipment for supplying heat and coefficients used in the correction arithmetic expression. To set the value of, it is necessary to make an adjustment work by data based on the past operation record. In addition, various conditions of heat source equipment for supplying heat are
Since it changes due to remodeling, etc.
You must change the constants and coefficients that have already been set.

[0008] Such a series of work is not used due to the complexity of the work because it is a manual work by an operator in actual operation, and the operator who operates the cooling and heating equipment actually has a past heat load record. It is often the case that the heat load of the next day is predicted by making an empirical decision based on the weather conditions of the next day and the current temperature.

However, it is difficult to reduce the burden on the operator in the operation of the building cooling and heating equipment which depends on the prediction of the heat load based on the experience of the operator. Further, if the conventional method is adopted as it is and the prediction of the heat load is continued, there is a problem that the operation efficiency of the heat source device is deteriorated due to a prediction error and the inexpensive nighttime power cannot be effectively used.

In addition, the amount of heat accumulated on the previous day is often insufficient or excessive, which causes a problem that heat cannot be appropriately supplied to the cooling and heating equipment on the day. Furthermore, it is known that the heat load fluctuates due to factors such as the day of the week and weather conditions such as the weather, but for example, the classification by day of the week may differ for each tenant. This is because the fixed holidays of tenants differ depending on the day of the week.

Therefore, since the conventional heat load prediction cannot perform statistical processing according to various classifications, when controlling the predicted number of heat source devices using a highly accurate predicted heat load value, the optimum predicted number is obtained. As a result, there is a problem that the running cost of the heat source device is increased due to an increase in the frequency of starting and stopping the heat source device and the life of the heat source device is shortened.

The present invention has been made in view of the above circumstances, and formulates a non-linear relationship between a heat load variation factor such as a day of the week, weather and the like and a heat load, and predicts a heat load predicted value for the next day without human intervention. An object is to provide a heat load prediction device that predicts with high accuracy.

[0013]

In order to solve the above-mentioned problems, the invention according to claim 1 is for transporting cold / hot heat produced by a target heat source facility to heat consuming equipment such as an air conditioner of a building. A heat load predicting device for predicting a heat load consumed on the next day by the heat consuming device, a heat load detecting means for detecting a heat load consumed by the heat consuming device, and a meteorological actual value such as temperature and a weather forecast Meteorological data input means for inputting a value, data storage means for storing the heat load, the actual weather value and the weather forecast value, and calendar data generating means for generating day of the week data. , The weather actual value, the weather forecast value, the heat load, the day of the week data, the heat load characteristic amount data of the morning and afternoon obtained from the actual value of the hour heat load, selectively using Load prediction model and time heat load pattern prediction model Based on the weighting coefficient and the prediction model stored in the prediction model storage means, the learning means for learning the weighting coefficient of the prediction model, the prediction model storage means for storing the weighting coefficient of the prediction model learned by the learning means, Time heat load pattern predicting means for predicting the time heat load pattern according to the above, and the data for sending the time heat load pattern predicted by the predicting means to the heat source equipment control means for controlling the target heat source equipment. And a thermal load predicting device provided with a data output means.

Next, in the invention corresponding to claim 2, when transporting cold / hot heat produced by the target heat source equipment to a heat consuming device such as an air conditioner of a building, the heat consuming device consumes the heat on the next day. In a heat load predicting device for predicting heat load, a heat load detecting means for detecting a heat load consumed by the heat consuming device, and a meteorological data input means for inputting a meteorological actual value such as an air temperature and a meteorological forecast value. Data storage means for storing the heat load, the meteorological performance value and the weather forecast value, calendar data generating means for generating day of the week data, the meteorological performance value, the meteorological forecast value, The heat load, the day of the week
And learning means for learning the weighting factors of the daily heat load prediction model and the hourly heat load pattern prediction model by selectively using the morning and afternoon heat load feature amount data obtained from the actual values of the hourly heat load. , A prediction model storage means for storing the weighting coefficient of the prediction model learned by the learning means, and a day heat load for predicting the day heat load based on the weight coefficient stored in the prediction model storage means and the prediction model Prediction means,
A temporal heat load pattern predicting means for predicting a temporal heat load pattern based on the weighting coefficient stored in the predictive model storage means and the predictive model, and a variable load amount due to heat consumption of the heat consuming equipment. Heat consumption equipment fluctuating load amount calculating means to be obtained, and time heat load prediction for obtaining a heat load prediction value in hour units from the daily heat load, the temporal heat load pattern and the fluctuating load quantity of the heat consumption equipment predicted by the predicting means. Means and sends the predicted heat load value in units of time to the heat source equipment control means for controlling the target heat source equipment.

Further, the invention according to claim 3 is a new model as a prediction model learning means and a solar heat load prediction means.
This network consists of a laural network.
The actual weather value, the weather forecast value and the daily heat load are input to the input layer of the network, the weighting coefficient between the neurons of each layer is corrected by the backpropagation method, and the weighting coefficient obtained by this correction is predicted. Calendar data stored in the model storage means and generated from the calendar data generating means.
Selecting a weighting factor of the neural network from the prediction model storage means based on the data, and using the selected weighting factor and the meteorological forecast value to predict the heat load for one forecast day. Is.

Further, in the invention corresponding to claim 4, the predictive model learning means and the time heat load pattern predicting means are:
A neural network is constructed, and the actual weather value, the weather forecast value, and the heat load in units of time are input to the input layer of this neural network, and the weighting factor between the neurons of each layer is backpropagated. And the weighting factor obtained by this correction is stored in the prediction model storage means, and based on the calendar data generated from the calendar data generation means, the prediction model storage means stores the neural network. Selecting a weighting coefficient for the time, predicting the time heat load pattern on the forecast day using the selected weight coefficient and the weather forecast value, and calculating the time heat load pattern on the forecast day The estimated value of the heat load in units of hours is estimated by apportioning the daily heat load estimated amount of the day obtained by the heat load estimation means.

[0017]

Therefore, according to the invention corresponding to claim 1, by taking the above means, the heat load consumed by the heat consuming equipment from the heat load detecting means and the weather data inputting means,
Meteorological actual values such as temperature and weather forecast values are stored in the data storage means. Thereafter, the predictive model learning means is a heat load characteristic in the morning and afternoon obtained from the actual value of the actual weather value, the weather forecast value, the heat load, and the time heat load based on the day of the week generated from the calendar data generating means. The weight data of the daily heat load prediction model and the time heat load pattern prediction model are learned by selectively using the quantity data, and stored in the prediction model storage means.

Here, the time heat load pattern prediction means is
Since the time heat load pattern is predicted based on the weighting coefficient and the prediction model stored in the prediction model storage means, and is sent to the heat source equipment control means for controlling the target heat source equipment via the data output means, A highly accurate time heat load pattern can be predicted without the need for manual calculation.

In the invention according to claim 2, the heat load consumed by the heat consuming equipment from the heat load detection means and the meteorological data input means, the actual weather value such as the temperature, and the weather forecast value are stored in the data storage means. To do. Thereafter, the predictive model learning means is a heat load characteristic in the morning and afternoon obtained from the actual value of the actual weather value, the weather forecast value, the heat load, and the time heat load based on the day of the week generated from the calendar data generating means. The weight data of the daily heat load prediction model and the time heat load pattern prediction model are learned by selectively using the quantity data, and stored in the prediction model storage means.

Here, the solar heat load predicting means predicts the solar heat load by using the weighting coefficient stored in the predictive model storage means and the prediction model, while the hourly heat load pattern predicting means is the same. The time heat load pattern is predicted using the weighting coefficient stored in the prediction model storage means and the prediction model.

Thereafter, the time heat load predicting means, in addition to the daily heat load and the time heat load pattern predicted by each of the predicting means, also the fluctuation of the heat consuming equipment obtained by the heat consuming equipment fluctuation load amount calculating means. Since the heat load predicted value in units of time is calculated from the load amount and sent to the heat source equipment control means, it is possible to predict the heat load with high accuracy without requiring manual calculation. Also,
Even if the predicted heat load for the day gradually deviates, the prediction model is modified based on the latest heat load detection value, and the heat load is predicted in consideration of the fluctuating load amount of the heat consuming equipment. Therefore, highly accurate heat load prediction is possible.

[0022]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of the device of the present invention. In the figure, reference numeral 1 denotes a heat load predicting apparatus main body for predicting a heat load predictive value, and the heat load predictive value predicted by the apparatus main body 1 is sent to a heat source facility control means 2, where the heat load predicting value is predicted. A control signal is obtained based on the value and the measurement signal from the target heat source equipment 3, and the target heat source equipment 3 is controlled.

The target heat source equipment 3 means a heating / cooling equipment or the like composed of one or more of a heat pump, a boiler, a heat storage tank, etc., and the cold / hot heat produced here is the heat transport system 4. It is transported to a heat consuming device 5 such as an air conditioner installed in the building via the air conditioner and used for cooling and heating.

The heat load predicting apparatus main body 1 has a heat load detecting means 6 for detecting a heat load consumed by the heat consuming device 5, and a meteorological performance value such as a maximum temperature and a minimum temperature which are factors of fluctuation of the heat load. In addition to the meteorological performance value input means 7 for inputting data,
Similarly, the weather forecast value input means 8 for inputting the weather forecast value data such as the maximum temperature and the minimum temperature, which also cause the variation of the heat load, and the load variation calculation of the heat consuming device for obtaining the variable load amount due to the heat consumption of the heat consuming device The means 9 and the like are connected, and internally have the following configuration.

That is, the heat load predicting apparatus main body 1 stores the data input from the external means 6 to 9.
Data storage means 11, calendar data generation means 12 for generating calendar data such as the day of the week, and data required for heat load prediction from the data storage means 11 to load the prediction model. A predictive model learning means 13 for learning the coefficient,
A prediction model storage unit 14 for storing the weighting coefficient of the prediction model learned by the prediction model learning unit 13 is provided.

Further, the heat load predicting apparatus main body 1 has a day heat load predicting means 15 for predicting a day heat load based on the stored weight coefficient data and a prediction model, and a weight coefficient data stored similarly. -Time heat load pattern predicting means 16 for predicting the time heat load pattern based on the data and the prediction model,
The daily heat load amount and the hourly heat load pattern of the daily heat load predicting means 15
Time heat load predicting means for predicting an hourly heat load (time heat load) by taking in the time heat load pattern of the heat predicting means 16 and the fluctuating load quantity inputted from the heat consuming equipment fluctuating load quantity calculating means 9. 17, and the heat load predicted value obtained by the time heat load predicting means 17 is used as the heat source equipment control means 2
And a data output means 18 for sending the data to the data storage means 11 and storing it in the data storage means 11 as well.

Next, the operation of the apparatus configured as described above will be described. First, from the meteorological actual value input means 7 and the meteorological forecast value input means 8 etc., the meteorological actual value data such as the maximum temperature and the minimum temperature of the previous day and the meteorological forecast value data such as the maximum temperature and the minimum temperature of the next day are input. And is stored in the data storage means 11. Further, data necessary for learning the prediction model, such as the heat load, is stored in the data storage means 11. A. Learning operation of the prediction model by the prediction model learning means 13 The prediction model learning means 13 fetches the maximum temperature, the minimum temperature, and the heat load data stored in the data storage means 11, and further, the calendar data. The following processing is executed for each day of the week mode (for example, holiday, Saturday, weekday, etc.) generated by the generating means 12. A-1 Temperature data The data on the maximum and minimum temperatures are processed as follows.

[0028]

[Equation 1] Next, the deviation from the average value is calculated by the following arithmetic expression.

[0029]

[Equation 2] A-2 Heat load data On the other hand, data related to heat load is processed by the following method.

[0030]

(Equation 3) Next, the deviation from the average value is calculated by the following arithmetic expression.

[0031]

[Equation 4] A-3 Heat load feature amount The morning load feature amount R _sa (i) is obtained from the following arithmetic expression.

[0032]

(Equation 5) Here, qs (i, j): Time heat load performance [cal / h] at j o'clock on i day, j: 23 to 22, that is, 22:00 to 22:00 on the next day is set as one day, for example, 9 o'clock (j = What is the time heat load of 9) 8:
It means the cumulative heat load from 00 to 8:59. Similarly, the afternoon feature amount R _sp (i) of the heat load is obtained from the following arithmetic expression.

[0033]

(Equation 6) A-4 Learning of solar heat load prediction model by neural network The predictive model learning means 13 learns the weight coefficient of the solar heat load prediction model using the data obtained by the above processing. Here, the solar heat load prediction model is as shown in FIG. 2, and has a three-layer neural network structure including, for example, an input layer, an intermediate layer, and an output layer.

At this time, in the neural network,
Maximum temperature of the day stored in the data storage means 11,
Data such as the minimum temperature, the maximum temperature of the previous day, the minimum temperature, and the heat load of the previous day is input to the input layer, and the heat load of the day stored in the data storage means 11 is used as teaching data. ,
The back propagation method learns the weighting coefficient between neurons.

The weighting coefficient for this daily heat load prediction is as _follows : W _v12 (i, j): Weighting coefficient between input layer and middle layer W _v23 (i, k): Weighting coefficient between middle layer and output layer i, j, k: There are the number of input layers, intermediate layers, and output layers.

Here, the back propagation method is
This is a learning method in which the network error is propagated back from the output layer to the input layer in contrast to the neural network having a hierarchical structure like this prediction model. It is a method.

The weighting coefficient of the solar heat load prediction model learned in this way is stored in the prediction model storage means 14. A-5 Learning of time heat load pattern prediction model by neural network By the same method as A-4, a time heat load pattern prediction model of a neural network as shown in FIG. 3 is obtained. It is used to input data such as maximum temperature, minimum temperature of the day, maximum temperature of the previous day, minimum temperature, morning feature amount of the previous day, feature amount of the previous day afternoon, etc. to the input layer for the neural network. Then, by using the feature amount of the morning of the day and the feature amount of the afternoon of the day as teaching data, the weighting coefficient between neurons is learned by the back propagation method.

The weight coefficients for predicting the temporal heat load _pattern include W _p12 (i, j): weight coefficient between the input layer and the intermediate layer W _p23 (i, k): weight coefficient between the intermediate layer and the output layer i , J, k: There are numbers of input layers, intermediate layers, and output layers. Here, the weighting coefficient of the learned temporal heat load pattern prediction model is stored in the prediction model storage means 14. B. Operation of solar heat load predicting means The solar heat load predicting means 15 uses the solar heat load predicting model shown in FIG. 2, and the weight of the solar heat load predicting model shown in FIG. Using the coefficient, the maximum temperature, the minimum temperature of the previous day, the expected maximum temperature of the day, the expected minimum temperature, and the heat load of the previous day are input to the input layer of the neural network, and the heat load of the day is predicted. Further, the reverse operation given by the equation (6), that is,

[0039]

(Equation 7) And predict the solar heat load prediction value. About the operation by the C time heat load pattern prediction means This time heat load pattern prediction means 16 uses the time heat load pattern prediction model shown in FIG. 3 and is stored in the prediction model storage means 14. The time heat load pattern is predicted by using the weighting coefficient of the time heat load pattern prediction model shown in FIG. 3 and by processing as shown in FIG.

That is, in the input layer of the neural network, the maximum temperature of the previous day, the minimum temperature, the expected maximum temperature of the day, the expected minimum temperature, the characteristic amount of the morning of the previous day, the characteristic amount of the afternoon of the previous day, the heat of the previous day. Input the load, and the feature quantity on the morning of the day,
A feature amount in the afternoon of the day is predicted (S1).

Further, the characteristic amount of the past performance (i days, for example, 30 days) is read from the data storage means 11, and the most similar day among them, that is, the day on which I (i) shown in the following equation is the smallest 1 A day is selected as a similar day (S2).

[0042]

[Equation 8]

If the similar date is the kth day,
The time heat load pattern of the k-th day is predicted as the time heat load pattern of the current day (S3). This time heat load pattern
Is the proration date shown by the following formula.

[0044]

[Equation 9]

P _s (k, i): Time heat load pattern on the kth day (i = 23 hours to 22:00) q _s (k, i): Time heat load on the kth day i (i = 23 hours
22:00) [cal / h] D Operation of time heat load predicting means This time heat load predicting means 17 is the day heat load predicting means 15
Based on the daily heat load predicted value and time heat load pattern obtained by the time heat load pattern prediction means 16 and the variable load amount output from the heat consumption record movable schedule calculation means 9, The time heat load prediction value is calculated by an equation.

[0046]

[Equation 10]

Then, the time heat load prediction value obtained by the time heat load prediction means 17 is sent to the heat source device control means 2 via the data output means 18, where the current day based on the time heat load prediction value. The target heat source equipment 3 is controlled.

Therefore, according to the configuration of the above embodiment, the heat load consumed on the next day in the heat consuming equipment is predicted based on the neuro model, and the weight in this neuro model is the past heat load. Since the backpropagation method is used for learning based on the actual results, the meteorological data, and the day of the week mode, it is possible to always predict the heat load with high accuracy.

As the factor data for varying the consumed heat load, the weather forecast value of the next day is adopted, and these data can be easily obtained by a weather forecast or the like. In addition, by classifying the load data according to the actual heat load and the actual result of the factors that affect the heat load, it is possible to predict the heat load with a small error for the heat load that largely changes such as the day of the week. Further, the heat load can be predicted while considering the heat load that fluctuates due to the heat consuming equipment operation schedule.

In the above embodiment, the maximum temperature and the minimum temperature are used as the meteorological data, but in addition, data such as weather and humidity may be used or combined in accordance with the heat load model. . Similarly, the day of the week is used as a factor that affects the heat load, but other weather conditions such as season and weather may be used, or a combination thereof may be used.

In the present embodiment, data is classified for each day of the week mode to make a prediction. However, the classification according to the day of the week mode is not performed, and the prediction information is incorporated in the input information of the prediction model. You can

Further, in the prediction of the daily heat load, one day may be divided into night (22:00 to 8:00) and daytime (8:00 to 22:00) and the respective predictions may be made. Furthermore, other time zones may be used as the method of obtaining the feature amount in the temporal heat load prediction. Moreover, the feature amount is not limited to two in the morning and afternoon, and may be one or three or more.

[0053]

As described above, according to the present invention, the non-linear relationship between the heat load fluctuation factors such as the day of the week and the weather and the heat load is formulated so that the heat load of the next day can be obtained without human intervention. Can predict the predicted value.

Further, since the predicted heat load value of the next day consumed by the heat consuming device is obtained according to the predictive model having the learning function, the predicted heat load value can be predicted with high accuracy, and the parameters of the model are consumed. Since the heat load to be estimated is estimated based on the factor data for changing the heat load, the heat load can be predicted with high accuracy at all times. Therefore, unlike the conventional case, the amount of heat supplied to the heat consuming device does not become insufficient or excessive, and the heat energy can be appropriately supplied.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing an embodiment of a heat load prediction device according to the present invention.

FIG. 2 is an explanatory diagram of a solar heat load prediction model using a neural network.

FIG. 3 is an explanatory diagram of a time heat load pattern prediction model using a neural network.

FIG. 4 is a diagram illustrating a procedure for predicting a temporal heat load pattern.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Heat load prediction apparatus main body, 3 ... Target heat source equipment, 5 ... Heat consuming equipment, 7 ... Meteorological actual value input means, 8 ... Meteorological forecast value input means, 9 ... Heat consuming equipment operating schedule calculation means, 1
DESCRIPTION OF SYMBOLS 1 ... Data storage means, 12 ... Calendar data generation means, 13 ... Prediction model learning means, 14 ... Prediction model storage means, 15 ... Sun heat load prediction means, 16 ... Hour heat load pattern prediction means , 17 ... Time heat load predicting means.

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI Technical display location G05B 13/04 9131-3H (72) Inventor Yukihiro Yamada 1-1-1 Shibaura, Minato-ku, Tokyo (72) Inventor Nobuto Fujita, No. 1 Toshiba-cho, Fuchu-shi, Tokyo Inside Toshiba Fuchu factory

Claims (4)

[Claims] 1. A heat load predicting device for predicting a heat load consumed on the next day in a heat consuming device such as an air conditioner of a building when transporting cold and hot heat produced by a target heat source facility to the heat consuming device. A heat load detecting means for detecting a heat load consumed by the heat consuming device; a meteorological data input means for inputting a meteorological result value such as an air temperature and a weather forecast value; the heat load, the meteorological result value and Data storage means for storing the weather forecast value, calendar data generating means for generating day of the week data, the meteorological performance value, the meteorological forecast value, the heat load, the day of the week data, A learning means for learning the weighting factors of the daily heat load prediction model and the hourly heat load pattern prediction model by selectively using the morning and afternoon heat load feature amount data obtained from the actual values of the hourly heat load. , Predictions learned by this learning method Prediction model storage means for storing the weighting coefficient of the model, and time heat load pattern prediction means for predicting the time heat load pattern based on the weighting coefficient stored in the prediction model storage means and the prediction model. A data for sending the time heat load pattern predicted by the predicting means to the heat source equipment controlling means for controlling the target heat source equipment.
A heat load predicting device comprising: 2. A heat load predicting device for predicting a heat load consumed on the next day in a heat consuming device such as an air conditioner of a building when transporting cold and hot heat produced by a target heat source facility to the heat consuming device. A heat load detecting means for detecting a heat load consumed by the heat consuming device; a meteorological data input means for inputting a meteorological result value such as an air temperature and a weather forecast value; the heat load, the meteorological result value and Data storage means for storing the weather forecast value, calendar data generating means for generating day of the week data, the meteorological performance value, the meteorological forecast value, the heat load, the day of the week data, A learning means for learning the weighting factors of the daily heat load prediction model and the hourly heat load pattern prediction model by selectively using the morning and afternoon heat load feature amount data obtained from the actual values of the hourly heat load. , Predictions learned by this learning method Prediction model storage means for storing model weighting factors, solar heat load prediction means for predicting solar heat load based on the weighting factors stored in the prediction model storage means and the prediction model, and the prediction model storage means Time heat load pattern predicting means for predicting a time heat load pattern based on the weighting coefficient stored in the above and the prediction model, and heat consumption equipment fluctuation for obtaining a fluctuation load amount due to heat consumption of the heat consumption equipment Load amount calculation means, day heat load, time heat load pattern predicted by the prediction means
And a heat load predicting means for obtaining a heat load predictive value for each hour from the fluctuating load amount of the heat consuming equipment, and A heat load predicting device characterized by sending out. 3. The predictive model learning means and the solar heat load predicting means constitute a neural network, and
The actual weather value, the weather forecast value and the daily heat load are input to the input layer of the Lar network and the weighting coefficient between the neurons of each layer is corrected by the backpropagation method. The weighting coefficient obtained by this correction is calculated. The weighting coefficient of the neural network is selected from the prediction model storage means based on the calendar data generated by the calendar data generation means and stored in the prediction model storage means. The heat load for one forecast day is predicted using the weighting coefficient and the weather forecast value.
Alternatively, the heat load prediction device described in 2. 4. The prediction model learning means and the time heat load pattern prediction means constitute a neural network,
The actual weather value, the weather forecast value, and the heat load in units of time are input to the input layer of this neural network, and the weighting coefficient between the neurons of each layer is corrected by the backpropagation method. A weighting coefficient is stored in the prediction model storage means, and a weighting coefficient of the neural network is selected from the prediction model storage means based on calendar data generated from the calendar data generation means, Using the selected weighting factor and the weather forecast value, the time heat load pattern of the forecast day is predicted, and the time heat load pattern of this forecast day is obtained on the day of the day obtained by the day heat load forecasting means. By apportioning by the predicted amount of heat load,
The heat load prediction device according to claim 1 or 2, which predicts a heat load prediction value in units of hours.