JP7650740B2 - Electricity demand forecasting device and electricity demand forecasting method - Google Patents

Description

The present invention relates to an electricity demand forecasting device and an electricity demand forecasting method.

Technologies have been known for some time now that predict the amount of electricity that will be needed in the future in order to ensure a stable supply of electricity, reduce power generation costs, and further reduce the burden on the environment. It is also known that the electricity demand to be predicted fluctuates due to the influence of natural factors such as weather. In recent years, electricity demand prediction models have been developed using mathematical methods based on various data on such natural factors.

For example, Patent Document 1 discloses a technique for predicting power demand on a target day by utilizing the correlation of the shape of annual fluctuations in past monthly power consumption. Patent Document 2 discloses a technique for predicting power demand on a target day by using a neural network to build a prediction model using actual data on past power demand and meteorological information.

Patent No. 4261471 Patent No. 3707589

However, with conventional technology, the forecast period for electricity demand was relatively short, making it difficult to forecast electricity demand for relatively long periods, up to one month or more in advance.

The present invention has been made to solve the above-mentioned problems, and aims to predict electricity demand over a relatively long period of time.

In order to solve the above-mentioned problems, the power demand forecasting device of the present invention includes an extraction unit that extracts a power demand predicted for a target period from an inferred value of power demand for a first observation period among past observation periods based on a probabilistic weather forecast value compared with a normal year value, and a presentation unit that presents the extracted predicted power demand, the inferred value being a value inferred for a first observation value, which is an observed value of weather for the first observation period, based on an inference model constructed in advance.

The power demand forecasting device according to the present invention may further include an inference unit that inputs the first observation value observed during the first observation period into the inference model constructed by learning the relationship between the second observation value, which is a meteorological observation value during a second observation period among the past observation periods, and the actual measured value of power demand, and calculates the inference model to obtain the inferred value of the power demand during the first observation period, and the inferred value obtained by the inference unit may be input to the extraction unit.

The power demand forecasting device according to the present invention may further include a learning unit that learns the relationship between the second observation value and the actual measurement value during the second observation period to construct the inference model, the second observation period being set according to information relating to the seasonal division of the target period, and the inference unit may read out the inference model from the learning unit and perform calculations of the inference model.

In addition, in the power demand forecasting device according to the present invention, the observed weather values are observed temperature values divided into a plurality of classes according to whether the values are high or low relative to the normal value, which is the average value of temperatures during the past observation period, and the probabilistic weather forecast values compared to the normal value are expressed as the probability that the temperature distribution related to the forecast falls into each of the plurality of classes, and the extraction unit may extract the predicted power demand using an average value obtained by weighting the inferred value corresponding to the possible class of the first observed value during the first observation period by the probability of each of the plurality of classes indicated by the forecast value.

In order to solve the above-mentioned problems, the power demand forecasting method according to the present invention includes an extraction step of extracting a power demand predicted for a target period from an inferred value of power demand for a first observation period among past observation periods based on a probabilistic weather forecast value compared with a normal year value, and a presentation step of presenting the extracted predicted power demand, wherein the inferred value is a value inferred for a first observation value, which is an observed value of weather for the first observation period, based on an inference model constructed in advance.

The power demand forecasting method according to the present invention may further include an inference step of inputting the first observation value observed during the first observation period into the inference model constructed by learning the relationship between the second observation value, which is a meteorological observation value during a second observation period among the past observation periods, and the actual measured value of power demand, and calculating the inference model to obtain the inferred value of the power demand during the first observation period, and the inferred value obtained by the inference step may be used in the extraction step.

The power demand forecasting method according to the present invention may further include a learning step of learning the relationship between the second observation value and the actual measurement value during the second observation period to construct the inference model, the second observation period being set according to information relating to the seasonal division of the target period, and the inference step may read out the inference model constructed in the learning step from a storage unit and perform calculations on the inference model.

According to the present invention, predictions are made based on probabilistic weather forecast values compared with average values, making it possible to forecast electricity demand over a relatively long period of time.

FIG. 1 is a block diagram showing a configuration of a power demand forecasting system including a power demand forecasting device according to a first embodiment of the present invention. FIG. 2 is a block diagram illustrating a hardware configuration of the power demand prediction device according to the first embodiment. FIG. 3 is a diagram for explaining an example of a long-term forecast input to the power demand forecasting device according to the first embodiment. FIG. 4 is a diagram for explaining an example of weather data used in the power demand forecasting device according to the first embodiment. FIG. 5 is a flowchart for explaining the operation of the power demand forecasting device according to the first embodiment. FIG. 6 is a diagram for explaining an example of a prediction result of power demand presented by the power demand prediction device according to the first embodiment. FIG. 7 is a diagram for explaining an example of a prediction result of power demand presented by the power demand prediction device according to the first embodiment. FIG. 8 is a block diagram showing a configuration of a power demand forecasting system including a power demand forecasting device according to the second embodiment. FIG. 9 is a flowchart for explaining the operation of the power demand forecasting device according to the second embodiment.

A preferred embodiment of the present invention will be described in detail below with reference to Figures 1 to 9. In this embodiment, the observed and forecasted weather values will be described as being the observed and forecasted temperature values.

[First embodiment]
1 is a block diagram showing a configuration of an electric power demand forecasting system including an electric power demand forecasting device 1 according to a first embodiment of the present invention. The electric power demand forecasting system according to the present embodiment predicts electric power demand for a relatively long period of time, up to one month or more in advance, based on weather forecast values, which are probabilistic long-term weather forecasts compared with normal values.

As shown in FIG. 1, the electricity demand forecasting system includes a weather forecast data storage unit 10, a weather data storage unit 11, and an electricity demand forecasting device 1. The electricity demand forecasting device 1 is connected to the weather forecast data storage unit 10 and the weather data storage unit 11 via a network NW. The weather forecast data storage unit 10 and the weather data storage unit 11 are, for example, databases provided in an external weather information system.

The power demand forecasting device 1 obtains from the weather forecast data storage unit 10 a probabilistic weather forecast that is compared with the average value for the target period for which the power demand is forecast. The weather forecast stored in the weather forecast data storage unit 10 is a long-term forecast of the temperature at the location or geographical area of the facility to be forecasted. For example, a three-month seasonal forecast issued by the Japan Meteorological Agency can be used.

The power demand forecasting device 1 also acquires meteorological observation values (first observation values) for a set first observation period from the past observation periods from the meteorological data storage unit 11. The meteorological observation values stored in the meteorological data storage unit 11 are temperature data observed at the location or geographical area of the facility to be predicted. For example, temperature observation data collected by AMeDAS (Automated Meteorological Data Acquisition System) for the past 30 years can be used as the first observation values for the first observation period.

[Functional blocks of the power demand forecasting device]
The power demand prediction device 1 includes an extraction unit 12, an inference unit 13, a presentation unit 14, and a first storage unit 15. The power demand prediction device 1 is installed, for example, in a facility whose power demand is to be predicted, and predicts a relatively long-term power demand in the facility up to three months ahead.

The extraction unit 12 extracts the power demand predicted for the target period from the inferred value of the power demand for the first observation period of the past observation periods based on the probabilistic weather forecast values compared with the average year value. The extraction unit 12 extracts the predicted power demand using the weather forecast values acquired from the weather forecast data storage unit 10. As described above, in this embodiment, the inferred values for the past 30 years are used as the inferred values for the first observation period of the past.

The probabilistic weather forecast values compared to normal values used by the extraction unit 12 are long-term forecasts such as one-month forecasts or three-month forecasts of temperature at the location or geographical region of the facility to be predicted. The forecast period of the long-term forecast corresponds to the length of the forecast period for which the electricity demand is to be predicted. For example, in this embodiment, the extraction unit 12 performs extraction processing based on a three-month forecast of temperature to extract electricity demand for the next three months.

Figure 3 is a schematic diagram showing an example of a probabilistic weather forecast value compared to a normal value used by the extraction unit 12. (a), (b), and (c) of Figure 3 show a three-month temperature forecast for each month. The probabilistic temperature forecast value compared to the normal value is divided into multiple classes according to the distribution of the forecasted temperatures relative to the normal value. The normal value is the average of the observed temperature data for the past 30 years. Furthermore, the temperature is divided into three classes relative to this normal value, namely, "high" which is higher than the normal value, "average" which is the same as the normal value, and "low" which is lower than the normal value.

Figure 3 (a) to (c) show the probability that temperatures for each prefecture one month, two months, and three months ahead will be in one of three categories - "high," "average," or "low" - compared to the average temperature for that month using different shades of color.

As shown in Figure 3(a), for example, in Hokkaido, the probability that the temperature forecast value one month ahead will be lower than the average is 10%, that it will be normal to the average is 30%, and that it will be higher than the average is 60%. Also, as shown in Figure 3(a), in each prefecture of Honshu one month ahead, the probability that it will be lower than the average is 20%, that it will be normal to the average is 40%, and that it will be higher than the average is 40%. Similarly, Figure 3(b) shows the probability that the temperature two months ahead will be in the "high", "average", and "low" classes, and Figure 3(c) shows the probability that the temperature three months ahead will be in the "high", "average", and "low" classes.

In this way, the extraction unit 12 uses as input the probability value of the temperature forecast for three months ahead in the area of the facility to be predicted. For example, in the example of the three-month temperature forecast shown in (a) to (c) of Figure 3, consider a case where electricity demand is predicted for a facility located in Hokkaido for three months ahead from March 2021. In this case, the extraction unit 12 uses as input the temperature forecast values for April 2021 "Low: Average: High (%): 10: 30: 60", the temperature forecast values for May 2021 "Low: Average: High (%): 20: 40: 40", and the temperature forecast values for June 2021 "Low: Average: High (%): 20: 30: 50".

Returning to FIG. 1, the extraction unit 12 extracts the power demand predicted for the prediction period using an inferred value of the power demand that has been inferred in advance for the temperature observation data (first observation value) for the past 30 years. In this embodiment, the extraction unit 12 uses the inferred value of the power demand for the past 30 years obtained by the inference unit 13 described below.

The inference unit 13 uses a pre-constructed inference model 15a to determine an inferred value of power demand for meteorological observation data over the past 30 years. As will be described in detail later, the inference model 15a is a model constructed by pre-learning the relationship between the observed temperature data (second observed value) during a set second observation period among past observation periods and the actual measured value of power demand. The inference unit 13 inputs meteorological values observed over the past 30 years into the inference model 15a and calculates the inference model 15a to determine an inferred value of power demand over the past 30 years.

More specifically, the inference unit 13 provides the inference model 15a with the temperature observation data for the past 30 years acquired from the weather data storage unit 11 of an external weather information system as unknown input data, and determines the value of the power demand to be inferred for the past 30 years in the facility to be predicted. The inference model 15a is constructed in advance, for example, on an external server by learning the relationship between past temperatures and power demand, and is stored in the first storage unit 15.

The presentation unit 14 presents the power demand predicted for the target period extracted by the extraction unit 12. For example, the presentation unit 14 displays the results of the power demand prediction for a set period, such as one month, on the display device 107 provided in the power demand prediction device 1.

The first memory unit 15 stores average values of temperature and values related to class classification that are used when the extraction unit 12 extracts the value of the predicted electricity demand. The first memory unit 15 also stores a pre-constructed inference model 15a that is used by the inference unit 13 for the inference process. The first memory unit 15 stores actual measured values of electricity demand at the facility to be predicted. Furthermore, the first memory unit 15 stores information related to data processing of the prediction results when the presentation unit 14 displays the predicted electricity demand.

Here, the process in which the extraction unit 12 extracts the predicted power demand for the facility to be predicted for the next three months (April to June 2021) will be described in more detail with reference to FIG. 4 as an example.

In this embodiment, the extraction unit 12 extracts the average value of power demand taking into account probability, that is, the value of power demand predicted for the prediction period from the inferred value of power demand over the past 30 years using expected value calculation.

Data a in FIG. 4 shows the observed data of temperature for the past 30 years acquired from the weather data storage unit 11 and the corresponding inferred values of power demand. Specifically, in the example of data a in FIG. 4, the inferred values of power demand arranged in ascending order of temperature in April for the past 30 years (1991 to 2020) are shown. As described above, in this embodiment, temperatures are classified into three classes "high", "average", and "low" in comparison with the average value. In the example of data a in FIG. 4, the temperatures for the top 10 years are classified into the class "low", the temperatures for the middle 10 years are classified into the class "average", and the temperatures for the bottom 10 years are classified into the class "high". In addition, in data a in FIG. 4, the area surrounded by a dashed rectangle shows the inferred values of power demand associated with the three classes "high", "average", and "low" classifications of the average value of temperature.

Data b in FIG. 4 is one month's worth of long-term forecast values for temperatures up to three months ahead that the extraction unit 12 retrieves from the weather forecast data storage unit 10, as described in FIG. 3, and shows the temperature probabilities by class. That is, FIG. 4 shows a long-term forecast in which, for example, in the area of the facility to be predicted, the temperature one month ahead, i.e., in April 2021, has a 70% probability of being higher than average ("High: 70%), a 20% probability of being average ("Average: 20%), and a 10% probability of being lower than average ("Low: 10%).

The extraction unit 12 uses a certain inferred value corresponding to the probability of the long-term temperature forecast from among the inferred values of electricity demand based on temperature observation data from the past 30 years for calculating the expected value. More specifically, the extraction unit 12 uses a number of inferred values of electricity demand corresponding to the probability values of the long-term temperature forecast for April 2021 shown in data b of FIG. 4 ("High: 70%", "Average: 20%", "Low: 10%)) for calculating the expected value using formula c of FIG. 4.

More specifically, the extraction unit 12 adjusts the range of the inferred power demand values to be selected from each class division according to the magnitude of the probability value of each class indicated by the long-term temperature forecast. For example, when the probability value of the class "high" is the highest among the three probability values of the long-term temperature forecast, the extraction unit 12 selects the inferred power demand value with the higher temperature value within each of the three classes "high", "average", and "low" from each of the three classes "high", "average", and "low" and uses it for the expected value calculation. Therefore, in this case, among the inferred power demand values belonging to the class "high", the inferred power demand value associated with the higher temperature value is selected. Also, among the inferred power demand values belonging to the class "average", the inferred power demand value associated with the higher temperature value is selected. Similarly, among the inferred power demand values belonging to the class "low", the inferred power demand value associated with the higher temperature value is selected.

On the other hand, if the probability value of the "low" class is the highest among the probability values of the three classes of long-term temperature forecasts, the extraction unit 12 selects the inferred power demand value with the lower temperature value among the three classes from the inferred power demand values corresponding to the temperatures of the observation data from the past 30 years, and calculates the expected value. Therefore, in this case, the inferred power demand value with the lower associated temperature value is selected from the inferred power demand values belonging to the "high" class. Also, the inferred power demand value with the higher associated temperature value is selected from the inferred power demand values belonging to the "average" class. Similarly, the inferred power demand value with the lower associated temperature value is selected from the inferred power demand values belonging to the "low" class.

Using the example of Figure 4, the probability value of the long-term forecast of temperature indicated by data b is the highest at 70% for the "high" class. For this reason, as shown by the dotted arrow "7 years" in Figure 4, the inferred value of power demand for 7 years is selected from the data classified as "high" class included in data a, starting from the data with the highest temperature values. Also, as shown by the dotted arrow "2 years" in Figure 4, the inferred value of power demand for 2 years is selected from the data classified as "average" class included in data a, starting from the data with the highest temperature values. Similarly, as shown by the dotted arrow "1 year" in Figure 4, the inferred value of power demand for 1 year is selected from the data classified as "low" class, starting from the data with the highest temperature values.

In this way, the extraction unit 12 extracts the power demand predicted in the forecast target period by using the average value, that is, the expected value, of the inferred power demand corresponding to the possible classes of "high", "average", and "low" of the temperature observation data values over the past 30 years weighted by the probability of each of these three classes indicated by the long-term temperature forecast. As represented by formula c in FIG. 4, the predicted power demand is extracted by calculating the expected value = {[inferred value of power demand for 7 years selected from the class "high"] + [inferred value of power demand for 2 years selected from the class "average"] + [inferred value of power demand for 1 year selected from the class "low"] / [10 years]}. In addition, similar to the calculation of the expected value for April 2021 shown in FIG. 4, the expected values can be calculated for May and June 2021, respectively, and the power demand predicted in the forecast target period can be extracted.

As described above, the extraction unit 12 performs extraction processing including a process of adjusting the range of inferred values of power demand selected from each class division according to the magnitude of the probability value of each class indicated by the long-term temperature forecast, and a process of calculating an expected value based on the selected inferred values of power demand. The extraction unit 12 may calculate the expected value using the average value of the inferred values of power demand included in each class division.

In the above-mentioned extraction unit 12, a case has been exemplified in which a long-term forecast for up to three months ahead is used on a monthly basis for forecast data ranging from the first day to the last day of each month. However, the specific range of the forecast for one month is not limited to the first day to the last day of the month. For example, the range of forecast data for April may be from March 28th to April 27th. In this case, the extraction unit 12 can use, for example, past temperature observation data from March 28th to April 27th from 1991 to 2020 as observation data for the past 30 years, and extract the power demand predicted for March 28th to April 27th, 2021.

[Hardware configuration of the power demand forecasting device]
Next, an example of a hardware configuration for realizing the power demand prediction device 1 having the above-described functions will be described with reference to FIG.

As shown in FIG. 2, the power demand prediction device can be realized by, for example, a computer including a processor 102, a main memory device 103, a communication interface 104, an auxiliary memory device 105, and an input/output I/O 106 connected via a bus 101, and a program that controls these hardware resources. The power demand prediction device 1 can also include a display device 107 connected via the bus 101.

The main memory device 103 stores in advance programs that allow the processor 102 to perform various controls and calculations. The processor 102 and the main memory device 103 implement the functions of the extraction unit 12 and the inference unit 13 shown in FIG. 1.

The communication interface 104 is an interface circuit for networking the power demand prediction device 1 and various external electronic devices. The communication interface 104 may be configured to receive weather forecast data and past weather data used by the extraction unit 12 described in FIG. 1, and the inference model 15a used by the inference unit 13 from an external terminal via the network NW.

The auxiliary storage device 105 is composed of a readable/writable storage medium and a drive for reading and writing various information such as programs and data from the storage medium. The auxiliary storage device 105 can use semiconductor memory such as a hard disk or flash memory as the storage medium.

The auxiliary storage device 105 has a program storage area for storing the power demand forecasting program executed by the power demand forecasting device 1. The auxiliary storage device 105 realizes the first storage unit 15 described in FIG. 1. The auxiliary storage device 105 also stores information regarding the average temperature, the high and low temperature classifications compared to the average temperature, and information regarding the formula for calculating the expected value. Furthermore, it may have, for example, a backup area for backing up the above-mentioned data, programs, etc.

The input/output I/O 106 is composed of I/O terminals that input signals from external devices and output signals to external devices.

The display device 107 is configured with an organic EL display, a liquid crystal display, or the like. The display device 107 can also realize the presentation unit 14 described in FIG. 1.

The program stored in the program storage area of the auxiliary storage device 105 may be a program that is processed chronologically in accordance with the order of the power demand forecasting method described in this specification, or may be a program that is processed in parallel or at the required timing, such as when called. In addition, the program may be processed by one computer, or may be processed in a distributed manner by multiple computers.

[Operation of the power demand forecasting device]
5 is a flowchart showing the operation of the power demand forecasting device 1 according to the present embodiment. First, the extraction unit 12 inputs a probabilistic weather forecast value compared with an average value, which is acquired, for example, via the network NW from the weather forecast data storage unit 10 of an external weather information system (step S1). Specifically, the extraction unit 12 inputs a long-term forecast of temperature for the next three months acquired from the weather forecast data storage unit 10. Note that in step S1, the extraction unit 12 is not limited to inputting a forecast value acquired via the network NW, and may be configured to read and input a forecast value previously stored in a portable memory such as a USB memory.

Next, the inference unit 13 provides the temperature observation data from the past 30 years as unknown input to the inference model 15a read from the first memory unit 15, performs calculations according to the inference model 15a, and obtains an inferred value of electricity demand for the temperature observation data from the past 30 years (step S2).

Next, the extraction unit 12 uses the probability value of the long-term temperature forecast input in step S1 to calculate the expected value for the inferred value of the power demand obtained in step S2 (step S3). Specifically, if the probability values for each class for a certain month in the long-term temperature forecast input in step S1 are "high": 70%, "average": 20%, and "low": 10%, the extraction unit 12 calculates the average value of the inferred value of the power demand for the past 30 years obtained in step S2 weighted by each of the probabilities of "high": 70%, "average": 20%, and "low": 10%.

Furthermore, when calculating the expected value in step S3, if the probability value of the long-term temperature forecast is the highest among the three classes of "high", "average", and "low", the extraction unit 12 can select the inferred value of the electricity demand for each class from the data for the higher temperature within each of these classes of "high", "average", and "low".

In addition, when the probability value of the long-term temperature forecast is the highest for the "low" class among the three classes of "high," "average," and "low," the extraction unit 12 can select the inferred value of the electricity demand for each class from the data for the lower temperature within each of these classes of "high," "average," and "low."

In this way, the extraction unit 12 adjusts the range of inferred values of electricity demand selected from each class interval depending on the magnitude of the probability value of each class indicated by the long-term temperature forecast.

Next, the extraction unit 12 obtains the power demand predicted for the prediction period from the expected value calculated in step S3 (step S4). Specifically, the extraction unit 12 can calculate the expected value in step S3 according to the forecast period related to the long-term forecast. For example, if the long-term forecast is a three-month forecast on a monthly basis, the extraction unit 12 can calculate the expected value in step S3 for each month and obtain the power demand for up to three months in advance.

Next, the presentation unit 14 presents the power demand forecast result obtained in step S4 on the display device 107 (step S5).

Figures 6 and 7 are diagrams showing examples of the results of the electricity demand forecast presented by the presentation unit 14. In area d of Figure 6, for each of the forecast target periods "April," "May," and "June," the "Forecast results (kWh)," "Last year's actual results (kWh)," and "Compared to last year" are displayed in association with each other.

Figure 7 shows an example of the forecast results of Figure 6 displayed by the display device 107 as a monthly bar graph. The horizontal axis of Figure 7 indicates the forecast target periods of "April 2021," "May 2021," and "June 2021," and the vertical axis indicates the amount of electricity [kWh]. Data e of the bar graph represents the actual value of electricity demand last year, and data f represents the value of electricity demand in the forecast results. As shown in Figure 7, by displaying electricity demand for the next three months in a form that can be compared with the actual value of the previous year, it becomes easier to grasp long-term electricity demand.

The presentation unit 14 can further display the predicted power demand in a graph, separated for each individual section or multiple sections included in the facility to be predicted. Alternatively, the presentation unit 14 can display the predicted power demand at multiple locations in a graph as forecast data for a wider area.

In the embodiment described above, the inference unit 13 reads out the pre-constructed inference model 15a from the first storage unit 15 and performs the inference process. However, the inference process by the inference unit 13 may be configured to be performed by an external server. In this case, the extraction unit 12 obtains the inferred value of the power demand from the external server via the network NW. In addition, in this case, the inference model 15a is stored in the external server.

As described above, the power demand forecasting device 1 according to the first embodiment predicts power demand based on a long-term probabilistic weather forecast compared with normal values, making it possible to forecast power demand over a relatively long period of time.

In addition, the power demand forecasting device 1 uses an average value of the inferred power demand over the past 30 years, which corresponds to the possible classes of temperature observation data over the past 30 years, weighted by the probability of each of the multiple classes indicated by the long-term forecast, making it easier to forecast power demand for a relatively long period of time, up to one month or more in advance.

[Second embodiment]
Next, a second embodiment of the present invention will be described. In the following description, the same components as those in the first embodiment are given the same reference numerals, and the description thereof will be omitted.

In the first embodiment, the extraction unit 12 acquires an inferred value of the power demand for a first observation period among past observation periods, that is, for the past 30 years, which is inferred in advance by the inference unit 13. Also, the inference unit 13 reads out an inference model 15a constructed in advance from the first storage unit 15 and performs inference processing. In contrast, in the second embodiment, a learning unit 16 is further provided, and performs learning processing of the inference model 15a.

As shown in FIG. 8, the power demand forecasting system according to this embodiment includes a power demand forecasting device 1A, a weather forecast data storage unit 10, a weather data storage unit 11, and a learning data storage unit 17.

[Functional blocks of the power demand forecasting device]
8, the power demand forecasting device 1A according to the present embodiment includes an extraction unit 12, an inference unit 13, a presentation unit 14, a first storage unit 15, and a learning unit 16. The power demand forecasting device 1A according to the present embodiment differs in configuration from the power demand forecasting device 1 according to the first embodiment in that the power demand forecasting device 1A according to the present embodiment includes the learning unit 16. The following description will focus on the configuration that differs from the first embodiment.

The learning unit 16 learns the relationship between the second observation value, which is the observation value of the temperature in a second observation period that is set in advance among the past observation periods, and the actual measured value of the power demand in the second observation period, and constructs the inference model 15a. The second observation period is set according to information on the seasonal division of the prediction target period, and specifically, is set according to the season and length of the prediction target period. For example, the learning unit 16 can use the observation data of the temperature for five months or one year of the previous year and the actual measured value of the power demand at the prediction target facility as learning data according to the season and length of the prediction target period. In addition, when the extraction unit 12 predicts the power demand from April to June 2021, the observation data of the temperature and the actual measured value of the power demand for five months of the previous year from March to July 2020 can be used as learning data.

Here, we will explain the second past observation period, which is the period from which learning data is acquired. As described above, the first past observation period is the period for which the inference unit 13 performs inference processing to obtain an inferred value of power demand, and in this embodiment, inferred values are obtained for temperature observation data from the past 30 years. On the other hand, the second past observation period from which learning data is acquired is set according to the season and length of the prediction target period as described above, and can be a past period that does not overlap with the first past observation period.

Specifically, the second past observation period may not overlap with the first past observation period, may be partly included in the first past observation period, or may be included in the first past observation period. For example, when predicting electricity demand in 2025, the 30 years from 1991 to 2020 may be used as the first observation period, and the temperature observation data for 2024 may be used as the second observation period related to the learning data.

The learning unit 16 acquires learning data for constructing the inference model 15a from the learning data storage unit 17, and performs a learning process using a known machine learning algorithm. In addition, each time the past temperature observation data used as learning data is updated in the weather information system, the learning unit 16 can re-learn and construct the inference model 15a based on the updated weather data.

The learning data storage unit 17 stores learning data used by the learning unit 16 to construct the inference model 15a. More specifically, the learning data storage unit 17 can store learning data consisting of observed data of temperature in the previous year and actual measured values of electricity demand in the facility to be predicted in the previous year.

[Operation of the power demand forecasting device]
Next, the operation of the power demand forecasting device 1A having the above-mentioned configuration will be described with reference to the flowchart in Fig. 9. First, the extraction unit 12 inputs a probabilistic weather forecast value compared with an average value, which is acquired from the weather forecast data storage unit 10 via the network NW (step S1). Specifically, the extraction unit 12 inputs a three-month forecast of temperature acquired from the weather forecast data storage unit 10. Note that in step S1, the extraction unit 12 is not limited to inputting a forecast value acquired via the network NW, and may be configured to read and input a forecast value stored in a portable memory such as a USB memory.

Next, the learning unit 16 obtains the learning data from the learning data storage unit 17 and constructs the inference model 15a (step S20). Specifically, the learning unit 16 uses a known machine learning algorithm to learn the relationship between the temperature observed in the set second observation period and the actual measured value of the power demand during that period. For example, if the period for predicting the power demand is three months, the learning unit 16 can use learning data consisting of the observation data of the temperature for five months in the previous year and the actual measured value of the power demand for those five months. The inference model 15a constructed by the learning unit 16 is stored in the first storage unit 15.

Next, the inference unit 13 reads out the inference model 15a constructed in step S20 from the first storage unit 15, and obtains an inferred value of the power demand (step S2). More specifically, the inference unit 13 provides the inference model 15a with the observation data of the temperature for the past 30 years as unknown input, performs a calculation according to the inference model 15a, and obtains an inferred value of the power demand for the past 30 years.

Next, the extraction unit 12 uses the probability value of the long-term forecast of temperature input in step S1 to calculate an expected value for the inferred value of the power demand obtained in step S2 (step S3). Next, the extraction unit 12 obtains the power demand predicted in the target forecast period from the expected value calculated in step S3 (step S4). Specifically, the extraction unit 12 can calculate the expected value in step S3 according to the forecast period related to the long-term forecast. For example, if the long-term forecast is a three-month forecast every month, the extraction unit 12 can calculate the expected value in step S3 for each month's worth of data and obtain the power demand predicted for up to three months ahead.

Next, the presentation unit 14 causes the display device 107 to present the power demand forecast result obtained in step S4 (step S5).

As described above, according to the second embodiment of the power demand forecasting device 1A, the learning unit 16 learns the relationship between past temperature observation data and past actual measured values of power demand to construct the inference model 15a, so that the inference model 15a can be reconstructed each time the past temperature observation data is updated, enabling more accurate long-term forecasting of power demand.

In the embodiment described above, all the functional units are provided in a single power demand prediction device 1. However, the functional units of the power demand prediction device 1 can be realized as a single computer or can be distributed across a network.

In the embodiment described above, the power demand forecasting device 1 uses a three-month forecast of monthly temperatures as a long-term forecast. However, the long-term forecast is not limited to a three-month forecast, and a long-term forecast for other periods, such as a one-month forecast, may be used. The one-month forecast may also include forecasts for each week. For example, the one-month forecast may be forecast data having forecast values for the first week, second week, and third and fourth weeks. In this case, the extraction unit 12 of the power demand forecasting device 1 can calculate expected values for the first week, second week, and third and fourth weeks according to the forecast period for the long-term forecast, and obtain the predicted power demand.

The above describes the embodiments of the power demand forecasting device and power demand forecasting method of the present invention, but the present invention is not limited to the described embodiments, and various modifications that a person skilled in the art can imagine can be made within the scope of the invention described in the claims.

1...power demand forecasting device, 10...weather forecast data memory unit, 11...weather data memory unit, 12...extraction unit, 13...inference unit, 14...presentation unit, 15...first memory unit, 15a...inference model, 101...bus, 102...processor, 103...main memory unit, 104...communication interface, 105...auxiliary memory unit, 106...input/output I/O, 107...display unit, NW...network.

Claims (2)

an extracting unit for extracting a predicted electricity demand for a target period from an inferred value of an electricity demand for a first observation period among past observation periods based on a probabilistic weather forecast value compared with a normal value;
a presentation unit that presents the extracted predicted power demand,
The inferred value is a value inferred for a first observation value, which is a meteorological observation value during the first observation period, based on a previously constructed inference model ;
The meteorological observation values are temperature observation values classified into a plurality of classes according to whether the values are high or low relative to the normal value, which is an average value of temperatures during the past observation period;
the forecast value of the probabilistic weather compared to the normal value is expressed as a probability that the distribution of temperature related to the forecast falls into each of the plurality of classes;
The extraction unit extracts the predicted power demand by using an average value obtained by weighting the inferred value corresponding to a class that the first observation value in the first observation period can take with a probability of each of the multiple classes indicated by the forecast value.
The power demand forecasting device according to the present invention is characterized in that: - an extraction step of extracting a predicted electricity demand for a period of interest from an inferred value of the electricity demand for a first observation period of past observation periods based on a probabilistic weather forecast value compared to a normal value;
and a presentation step of presenting the extracted predicted power demand,
The inferred value is a value inferred for a first observation value, which is a meteorological observation value during the first observation period, based on a previously constructed inference model ;
The meteorological observation values are temperature observation values classified into a plurality of classes according to whether the values are high or low relative to the normal value, which is an average value of temperatures during the past observation period;
the forecast value of the probabilistic weather compared to the normal value is expressed as a probability that the distribution of temperature related to the forecast falls into each of the plurality of classes;
The extraction step extracts the predicted power demand by using an average value of the inferred value corresponding to a class that the first observation value in the first observation period can take, weighted by a probability of each of the multiple classes indicated by the forecast value.
A method for forecasting electricity demand.