JP7330077B2 - POWER DEMAND PREDICTION SYSTEM AND CONTROL METHOD FOR POWER DEMAND PREDICTION SYSTEM - Google Patents

Description

The present invention relates to a power demand forecasting system and a control method for the power demand forecasting system.

With the diversification of price menus due to the liberalization of the electricity retail market that started in April 2016, there is an increasing demand for a highly accurate electricity demand forecasting system that is indispensable for creating optimal equipment operation plans for buildings. ing. Therefore, the development of a new power demand prediction system that utilizes artificial intelligence (AI) technology such as deep learning is active.

In the AI method, learning is performed using past performance data to create a model (learning model) for demand forecasting. Regarding this demand forecasting model, the applicant of the present application has already proposed a power demand forecasting system as shown in Patent Document 1 (unpublished at the time of filing of the present application).

Japanese Patent Application No. 2018-221456 Japanese Patent No. 6372684 Japanese Patent No. 5822207

In general, the prediction accuracy of the AI method depends on the number of data used for learning, so it is said that the larger the number of data, the better the accuracy. However, in the case of newly built buildings, the amount of data that can be used for learning is small, so the prediction accuracy tends to be poor. For this reason, it is said that the number of data for at least one year (spring, summer, autumn and winter) is required. "Transfer learning", in which a model trained in one area is used in another area, enables learning and prediction with a small amount of data. difficult to collect in

The present invention has been made in view of the above circumstances, and is a problem in the power demand forecasting by the AI method, which is a power demand forecasting system and power demand forecasting system that can improve the forecasting accuracy with a small amount of data in a newly built building. It is an object of the present invention to provide a control method for a prediction system.

In order to solve the above problems, one aspect of the present invention is a power demand prediction system for predicting the power demand of a building, comprising: a first prediction model created by an artificial intelligence method for calculating a predicted value of the power demand; and a second prediction model created by a statistical method without a learning function that calculates the predicted value of the power demand, and a first predetermined period after starting the prediction performed by the power demand prediction system a model generating unit that inputs actual data including actual values of power demand of the building, weather information, and calendar information to the first prediction model and the second prediction model, respectively, and generates a first model and a second model; , the performance data for a second predetermined period of time shorter than the first predetermined period and past from the time of determining the forecast model is input to each of the first model and the second model, and the output of both models Either one of the first model prediction value and the second model prediction value and the actual value of the second predetermined period are input to a predetermined formula, and the first output as a calculation result of the predetermined formula a model determination unit that compares the prediction accuracy of the first model and the prediction accuracy of the second model, and determines either one of the first model and the second model as the prediction model according to the comparison result; performance data of the building for a period longer than the second predetermined period and past a third predetermined period from the prediction target date for calculating the predicted value of the power demand, and the prediction target date a prediction server having a power demand prediction unit that inputs data consisting of weather information and calendar information into the prediction model and calculates a predicted value of power demand on the prediction target day. It is a power demand forecasting system.

Further, according to one aspect of the present invention, in the power demand forecasting system, the model determining unit predicts the second model when the forecasting accuracy of the second model is equal to or less than the forecasting accuracy of the first model. model, and if the prediction accuracy of the second model is greater than the prediction accuracy of the first model, the first model is determined as the prediction model.

Further, one aspect of the present invention is the power demand prediction system, wherein the third predetermined period is a period in which the prediction accuracy of the second model is greater than the prediction accuracy of the first model. The period is shorter than the period.

Further, according to one aspect of the present invention, in the power demand forecasting system, the model determining unit of the forecasting model determination server provides each of the first model and the second model with a period of time shorter than the first predetermined period of time. a predicted value acquisition unit that inputs the performance data for a past second predetermined period from the time the prediction model is determined, and acquires the first model predicted value and the second model predicted value that are the outputs of both models; , either the first model predicted value or the second model predicted value, and the actual value for the second predetermined period are input into a predetermined formula, and the prediction accuracy of the first model and the second model A prediction accuracy calculation unit that outputs each prediction accuracy as a calculation result of the predetermined formula, and the calculated prediction accuracy of the first model and the calculated prediction accuracy of the second model are compared, and according to the comparison result, the and a model comparison result output unit that outputs one of the first model and the second model as a prediction model.

Further, one aspect of the present invention is the power demand prediction system, wherein the power demand prediction unit of the prediction server calculates the predicted value of the power demand for a period longer than the second predetermined period. a data acquisition unit for acquiring data consisting of performance data of the building for a third predetermined period in the past from the prediction target date and weather information and calendar information for the prediction target date; and a power demand calculation unit that inputs to the prediction model and calculates a predicted value of power demand on the prediction target day.

Further, according to one aspect of the present invention, a first prediction model created by an artificial intelligence method, which calculates the predicted value of the power demand, and a statistical model without a learning function, which calculates the predicted value of the power demand a second forecast model created by the method, and a forecast model determination server that calculates the actual power demand of the building for a first predetermined period after the model generation unit starts the forecast performed by the power demand forecast system; a step of inputting performance data including values, weather information and calendar information into the first prediction model and the second prediction model respectively to generate the first model and the second model; The performance data for a period shorter than the first predetermined period and past the second predetermined period from the time of determining the prediction model is input to each of the model and the second model, and the first Either one of the model predicted value and the second model predicted value and the actual value for the second predetermined period are input to a predetermined formula, and the prediction of the first model output as the calculation result of the predetermined formula comparing the accuracy and the prediction accuracy of the second model, and determining either one of the first model and the second model as the prediction model according to the comparison result, wherein the prediction server the actual data of the building for a period longer than the second predetermined period and a past third predetermined period from the prediction target date for which the predicted value of the power demand is calculated, and the prediction target date; inputting data consisting of weather information and calendar information into the prediction model to calculate a predicted value of power demand for the prediction target day. .

According to the present invention, the power demand forecasting system determines the second model as the forecasting model according to the comparison result even when the actual data of the building in the first predetermined period is small. It is possible to improve the prediction accuracy on the prediction target day with a small number of data.

1 is a configuration diagram showing a configuration example of a power demand prediction system according to an embodiment of the present invention; FIG. FIG. 2 is a schematic diagram for explaining the operation of the power demand prediction system shown in FIG. 1; 2 is a flow chart showing an operation example of the power demand prediction system shown in FIG. 1; 2 is a diagram showing predicted values and actual values when the power demand prediction system shown in FIG. 1 uses a first prediction model; FIG. 2 is a diagram showing predicted values and actual values when the power demand prediction system shown in FIG. 1 uses a second prediction model; FIG.

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a configuration diagram showing a configuration example of a power demand prediction system according to one embodiment of the present invention. Moreover, FIG. 2 is a schematic diagram for explaining the operation of the power demand prediction system shown in FIG.
The power demand forecasting system 10 according to one embodiment of the present invention shown in FIG. 1 prepares two models, an AI method and a statistical method, as shown in FIG. This is a power demand forecasting system that selects a highly accurate model.
Here, in the prediction model determination server, the prediction model (first prediction model) by the AI method is a prediction model (second prediction model) by a statistical method, such as the model described in "Japanese Patent Application No. 2018-221456" uses the model described in “Japanese Patent Application No. 2014-74239 (Patent No. 6372684)”.
A forecasting model based on a statistical method extracts the weather information for the forecast target date from the most recent actual data (weather information, calendar information), and the actual values of electricity demand for days similar to the calendar information. A forecast of power demand is calculated by an informed weighted average (weighted average).

Here, FIG. 4 is a diagram showing predicted values and actual values when the power demand prediction system shown in FIG. 1 uses the first prediction model. FIG. 5 is a diagram showing predicted values and actual values when the power demand prediction system shown in FIG. 1 uses the second prediction model. In both figures, the horizontal axis is time (the number of frames in 30-minute units, with 48 frames per day), and the vertical axis is the actual and predicted power values. Fig. 4 is a diagram showing the actual and predicted values of power consumption by the AI method when learning data for the past four years and creating a learning model. Estimated values are shown. On the other hand, FIG. 5 is a diagram showing actual values and predicted values of electric power by a statistical method when the predicted values are calculated using data for the past two weeks. shows the actual and predicted values of As shown in FIG. 4, the prediction error (EEP; Expected Error Percentage) by the AI method is a prediction accuracy of 2.888% because the first prediction model created from the number of data for four years is used. On the other hand, the prediction error (EEP) by the statistical method uses the second prediction model created from the data of the past two weeks, so the prediction accuracy is 4.334%, compared to the prediction accuracy of 2.888% by the AI method. low value.
However, unlike the AI method, the statistical method does not have a learning function (no data is required for learning). Sufficient accuracy can be obtained with the number of data for the first predetermined period). Therefore, when the actual data (number of data used for learning) is small, the statistical method is more accurate than the AI method.
In addition, since the statistical method uses only the most recent past performance data to make predictions (similar to AI method learning, past years (after starting the prediction made by the power demand forecasting system, ) data is not used), it is possible to respond immediately even if the power demand pattern changes (different from one year ago) due to changes in facility operation methods, etc.

That is, as shown in FIG. 1, the prediction model determination server 20 in the power demand prediction system 10 is a first prediction model 111 (prediction model shown in FIG. 1) and a second prediction model 112 (prediction model 2 shown in FIG. 2) that is created by a statistical method that does not have a learning function and calculates a predicted value of power demand.
The predictive model determination server 20 also has a model generator 21 . After starting the prediction performed by the power demand prediction system 10, the model generating unit 21 generates actual data (data shown in FIG. 2) including the actual value of the power demand of the building in the first predetermined period, weather information, and calendar information. , to the first prediction model 111 and the second prediction model 112, respectively, and generate the first model 11 and the second model 12 in which the coefficients used in each model are changed corresponding to the first predetermined period. .

As shown in FIG. 2, both models are input with actual data (power demand, weather information, calendar information) for the most recent past (second predetermined period; for example, for one week) to make predictions.
That is, the prediction model determination server 20 has a prediction value acquisition unit 22 . The predicted value acquisition unit 22 inputs performance data for a past second predetermined period from when the prediction model is determined, which is shorter than the first predetermined period, to each of the first model 11 and the second model 12, and Obtain the first model predicted value and the second model predicted value, which are the outputs of both models.

As shown in FIG. 2, in order to compare prediction accuracy (for example, second predetermined period=once a week), for example, EEP shown in the following formula is calculated and a model with good EEP is adopted.

In the above formula (1), _yp is the predicted value, _yd is the actual value, and _{yd. max} is the maximum actual value, and n is the number of data.
That is, the prediction model determination server 20 has a prediction accuracy calculator 23 as shown in FIG. The prediction accuracy calculation unit 23 inputs either one of the first model prediction value or the second model prediction value and the actual value of the second predetermined period to the above formula (1) (predetermined formula), and calculates the first Each of the prediction accuracy of the model 11 and the prediction accuracy of the second model 12 is output as the calculation result of the above equation (1).
The predictive model determination server 20 also has a model comparison result output unit 24 . The model comparison result output unit 24 compares the calculated prediction accuracy of the first model 11 and the calculated prediction accuracy of the second model 12, and selects either the first model 11 or the second model 12 according to the comparison result. Output as a prediction model.
Specifically, when the prediction accuracy of the second model 12 is equal to or lower than the prediction accuracy of the first model 11, the model comparison result output unit 24 determines the second model 12 as the prediction model, and is greater than the prediction accuracy of the first model 11, the first model 11 is determined as the prediction model.

That is, as shown in FIG. 1, the prediction model determination server 20 in the power demand prediction system 10 includes a model determination unit including a prediction value acquisition unit 22, a prediction accuracy calculation unit 23, and a model comparison result output unit 24. has 30. The model determination unit 30 inputs performance data for a past second predetermined period from the time of determining the forecast model, which is shorter than the first predetermined period, to each of the first model 11 and the second model 12. Either one of the first model prediction value and the second model prediction value, which is the output of the model, and the actual value for the second predetermined period are input to the equation (1), and the result of the calculation of the equation (1) is output. The prediction accuracy of the first model 11 and the prediction accuracy of the second model 12 are compared, and one of the first model 11 and the second model 12 is determined as the prediction model according to the comparison result.

As shown in FIG. 2, the prediction server acquires actual data (power demand, weather information, calendar information) for the most recent past (third predetermined period: for example, one month) and weather information and calendar information for the prediction target day. .
That is, the prediction server 40 has a data acquisition unit 41 as shown in FIG. The data acquisition unit 41 acquires performance data of the building for a period longer than the second predetermined period and past a third predetermined period from the prediction target date for calculating the predicted value of power demand, and weather information and calendar for the prediction target date. Obtaining data consisting of information and If the third predetermined period is shorter than the first predetermined period during which the prediction accuracy of the second model 12 is greater than the prediction accuracy of the first model 11, the second model 12 is determined as the prediction model. Therefore, it is possible to improve the prediction accuracy on the prediction target day with a small number of data in a newly built building.

The prediction server, as shown in FIG. 2, inputs the obtained data into the prediction model determined above and performs prediction.
That is, the prediction server 40 has a power demand calculator 42 as shown in FIG. The power demand calculation unit 42 inputs the data acquired by the data acquisition unit 41 into the prediction model, and calculates the predicted value of the power demand on the prediction target day.
The prediction server 40 may also have a prediction result output unit 43 that outputs the prediction result. The prediction result output unit 43 outputs, for example, the predicted value of the power demand calculated by the power demand calculation unit 42 to a building operation management device (not shown in FIG. 1). The operation management device is, for example, the operation management device described in Japanese Patent No. 5822207, and is a device that makes an optimum equipment operation plan for a building. The prediction model determination server 20 takes in the actual value of the power demand consumed by the building from the operation management device. Also, the prediction result output unit 43 outputs the power demand prediction value to the operation management device.

That is, the prediction server 40 has a power demand prediction unit 50 including a data acquisition unit 41, a power demand calculation unit 42, and a prediction result output unit 43, as shown in FIG. The power demand forecasting unit 50 generates performance data of buildings for a third predetermined period past the prediction target date for which the predicted value of power demand is calculated, which is longer than the second predetermined period, weather information for the prediction target day, and Data consisting of calendar information and is input to the prediction model to calculate the predicted value of power demand on the day to be predicted.

Next, an operation example of the power demand prediction system shown in FIG. 1 will be described with reference to FIG. FIG. 3 is a flow chart showing an operation example of the power demand prediction system shown in FIG.
Actual values are taken in (step ST1).
Specifically, after starting the prediction performed by the power demand prediction system 10, the model generation unit 21 generates the actual data (see FIG. 2 data shown in ).

A first model and a second model are generated. (Step ST2).
Specifically, the model generation unit 21 inputs actual performance data to each of the first prediction model 111 and the second prediction model 112, and changes the coefficients used in each model corresponding to the first predetermined period. generate a first model 11 and a second model 12 that are

A first model predicted value and a second model predicted value are obtained (step ST3).
Specifically, the predicted value obtaining unit 22 obtains the results of the past second predetermined period from the time the prediction model is determined, which is shorter than the first predetermined period, for each of the first model 11 and the second model 12. Input the data and get the first and second model predictions which are the outputs of both models.

The prediction accuracy of the first model and the prediction accuracy of the second model are calculated (step ST4).
Specifically, the prediction accuracy calculation unit 23 inputs either one of the first model prediction value or the second model prediction value and the actual value of the second predetermined period to the above equation (1), and the first Each of the prediction accuracy of the model 11 and the prediction accuracy of the second model 12 is output as the calculation result of the above equation (1).

A comparison is made as to whether or not the prediction accuracy of the first model and the prediction accuracy of the second model satisfy "prediction accuracy of the first model≧prediction accuracy of the second model" (step ST5).
Specifically, the model comparison result output unit 24 compares the prediction accuracy of the first model and the prediction accuracy of the second model, and depending on the comparison result (step ST5-Yes, step ST5-No), the first model and the second model as a predictive model.

A second model is determined as a prediction model (step ST6).
Specifically, when the prediction accuracy of the second model 12 is equal to or lower than the prediction accuracy of the first model 11 (step ST5-Yes), the model comparison result output unit 24 determines the second model 12 as the prediction model (step ST6).

A first model is determined as a prediction model (step ST7).
On the other hand, when the prediction accuracy of the second model 12 is higher than the prediction accuracy of the first model 11 (step ST5-No), the model comparison result output unit 24 determines the first model 11 as the prediction model.

Acquire data consisting of past performance data, weather information for the forecast target day, and calendar information (step ST8).
Specifically, the data acquisition unit 41 obtains the actual data of the building for a third predetermined period past the prediction target date for calculating the predicted value of power demand, which is longer than the second predetermined period, and the prediction target date. Get data consisting of weather and calendar information for

The acquired data is input to the determined prediction model, and prediction is performed (step ST9).
Specifically, the power demand calculation unit 42 converts the data acquired by the data acquisition unit 41 into a prediction model (either the second model 12 determined in step ST6 or the first model 11 determined in step ST7). model), and calculate the predicted value of power demand on the target day.

As described above, the following can be expected by applying the power demand prediction system 10 of the present invention to buildings such as office buildings.
It is possible to prevent deterioration of prediction accuracy due to new buildings with little data and changes in power demand patterns.
In addition, since the accuracy of the building equipment operation plan depends on the prediction accuracy, even in the above case, it is possible to formulate a highly accurate equipment operation plan, which contributes to energy saving and power saving of the building.

The power demand prediction system 10 in the above-described embodiment may be realized by a computer. In that case, a program for realizing this function may be recorded in a computer-readable recording medium, and the program recorded in this recording medium may be read into a computer system and executed. It should be noted that the "computer system" referred to here includes hardware such as an OS and peripheral devices. The term "computer-readable recording medium" refers to portable media such as flexible discs, magneto-optical discs, ROMs and CD-ROMs, and storage devices such as hard discs incorporated in computer systems. Furthermore, "computer-readable recording medium" refers to a program that dynamically retains programs for a short period of time, like a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line. It may also include something that holds the program for a certain period of time, such as a volatile memory inside a computer system that serves as a server or client in that case. Further, the program may be for realizing a part of the functions described above, or may be capable of realizing the functions described above in combination with a program already recorded in the computer system. It may be implemented using a programmable logic device such as an FPGA (Field Programmable Gate Array).

Although the embodiment of the present invention has been described in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and design and the like are included within the scope of the gist of the present invention.

DESCRIPTION OF SYMBOLS 10... Electric power demand forecasting system 11... 1st model 12... 2nd model 20... Prediction model determination server 21... Model generation part 22... Prediction value acquisition part 23... Prediction accuracy calculation part 24... Model comparison Result output unit 30 Model determination unit 40 Prediction server 41 Data acquisition unit 42 Power demand calculation unit 43 Prediction result output unit 50 Power demand prediction unit 111 First prediction model 112 …Second Predictive Model

Claims (6)

A power demand forecasting system for predicting the power demand of a building,
A first prediction model created by an artificial intelligence method that calculates the predicted value of the power demand;
a second prediction model created by a statistical method without a learning function, which calculates the predicted value of the power demand;
After starting the prediction performed by the power demand prediction system, the actual data including the actual value of the power demand of the building in a first predetermined period, weather information and calendar information are applied to the first prediction model and the second prediction model. a model generation unit that inputs to each and generates a first model and a second model;
The performance data for a second predetermined period of time shorter than the first predetermined period and in the past from the time of determining the forecast model is input to each of the first model and the second model, and the output of both models is Either one of the first model prediction value and the second model prediction value and the actual value of the second predetermined period are input to a predetermined formula, and the first model output as the calculation result of the predetermined formula a model determination unit that compares the prediction accuracy of the model and the prediction accuracy of the second model, and determines either one of the first model and the second model as the prediction model according to the comparison result;
a predictive model determination server having
Performance data of the building for a period longer than the second predetermined period and a past third predetermined period from the prediction target date for calculating the predicted value of power demand, weather information and calendar information for the prediction target date; A power demand forecasting unit that inputs data consisting of
a prediction server having
A power demand forecasting system comprising: The model determination unit
if the prediction accuracy of the second model is equal to or less than the prediction accuracy of the first model, determining the second model as the prediction model;
if the prediction accuracy of the second model is greater than the prediction accuracy of the first model, determining the first model as the prediction model;
The power demand forecasting system according to claim 1, characterized by: The third predetermined period is a period shorter than the first predetermined period during which the prediction accuracy of the second model is greater than the prediction accuracy of the first model.
3. The power demand forecasting system according to claim 1 or 2, characterized in that: The model determination unit of the prediction model determination server,
The performance data for a second predetermined period of time shorter than the first predetermined period and in the past from the time of determining the forecast model is input to each of the first model and the second model, and the output of both models is a predicted value acquisition unit that acquires certain first model predicted values and second model predicted values;
Either the first model predicted value or the second model predicted value and the actual value of the second predetermined period are input into a predetermined formula, and the prediction accuracy of the first model and the prediction accuracy of the second model A prediction accuracy calculation unit that outputs each prediction accuracy as a calculation result of the predetermined formula;
A model comparison result of comparing the calculated prediction accuracy of the first model and the calculated prediction accuracy of the second model, and outputting one of the first model and the second model as a prediction model according to the comparison result. an output unit;
The power demand forecasting system according to any one of claims 1 to 3, characterized by comprising: The power demand forecasting unit of the forecasting server,
Performance data of the building for a period longer than the second predetermined period and a past third predetermined period from the prediction target date for calculating the predicted value of power demand, weather information and calendar information for the prediction target date; a data acquisition unit that acquires data consisting of
a power demand calculation unit that inputs the data acquired by the data acquisition unit into the prediction model and calculates a predicted value of power demand on the prediction target day;
The power demand forecasting system according to any one of claims 1 to 4, characterized by comprising: A control method for a power demand prediction system that predicts the power demand of a building, comprising:
A first prediction model created by an artificial intelligence method that calculates the predicted value of the power demand;
A prediction model determination server having a second prediction model created by a statistical method without a learning function that calculates the predicted value of the power demand,
After the model generation unit starts the prediction performed by the power demand prediction system, the actual data including the actual value of the power demand of the building in a first predetermined period, weather information and calendar information is generated by the first prediction model and inputting to each of the second predictive models to generate a first model and a second model;
A model determination unit inputs the performance data for a second predetermined period in the past from when the prediction model is determined, which is shorter than the first predetermined period, to each of the first model and the second model, and Either one of the first model prediction value and the second model prediction value, which are the outputs of both models, and the actual value for the second predetermined period are input to a predetermined formula, and output as the calculation result of the predetermined formula. comparing the prediction accuracy of the first model and the prediction accuracy of the second model, and determining either one of the first model and the second model as the prediction model according to the comparison result; ,
has
The prediction server
A power demand forecasting unit, which is a period longer than the second predetermined period and past a third predetermined period from the forecast target date for calculating the forecasted value of the power demand, and the actual data of the building and the weather on the forecast target day. A step of inputting data consisting of information and calendar information into the prediction model and calculating a predicted value of power demand on the prediction target day;
A control method for a power demand forecasting system, comprising:

Images