JP2024152838A - Electricity demand forecasting system and electricity demand forecasting method - Google Patents

Abstract

To provide a system which is effective for improvement of prediction accuracy of power demand.SOLUTION: A power demand prediction system 10 comprises: a data accumulation unit 114 which accumulates teacher data including result data of a fluctuation factor of power demand and result data of the power demand for each of a plurality of customer groups of a power selling company in a database; a model generation unit 116 which generates a demand prediction model 118 representing a relation between fluctuation factor data and power demand data on the basis of a plurality of pieces of teacher data accumulated in the database for each of the plurality of customer groups; an input information acquisition unit 112 which acquires prediction data of the fluctuation factor for each of the plurality of customer groups; an individual demand prediction unit 121 which calculates prediction data of the power demand on the basis of the prediction data of the fluctuation factor and the demand prediction model 118 for each of the plurality of customer groups; and a total demand prediction unit 122 which calculates prediction data of the total power demand to the power selling company on the basis of the prediction data of the power demand calculated for each of the plurality of customer groups.SELECTED DRAWING: Figure 2

Description

This disclosure relates to an electricity demand forecasting system and an electricity demand forecasting method.

Patent Document 1 discloses an electricity demand forecasting system that includes a model creation unit that generates a model for predicting electricity demand by performing machine learning using learning data in which actual electricity data, which is the actual electricity demand, is associated with actual fluctuation factor data, which is the actual fluctuation factors that can cause fluctuations in electricity demand, and a forecasting unit that forecasts electricity demand on a target forecast date using the model.

JP 2019-204459 A

This disclosure provides a system that is effective in improving the accuracy of electricity demand forecasts.

The electricity demand forecasting system according to one aspect of the present disclosure includes a data storage unit that stores, for each of a plurality of customer groups of an electricity seller, training data including actual data on fluctuation factors of electricity demand and actual data on electricity demand in a database; a model generation unit that generates, for each of a plurality of customer groups, a demand forecast model that represents the relationship between the fluctuation factor data and electricity demand data based on the plurality of training data stored in the database; an input information acquisition unit that acquires predicted data of the fluctuation factors for each of the plurality of customer groups; an individual demand forecasting unit that calculates predicted data of electricity demand based on the predicted data of the fluctuation factors and the demand forecasting model for each of the plurality of customer groups; and a total demand forecasting unit that calculates predicted data of total electricity demand for the electricity seller based on the predicted data of electricity demand calculated for each of the plurality of customer groups, and the model generation unit generates, for one customer group, an electricity demand forecasting model that represents the relationship between the fluctuation factor data including one fluctuation factor and the electricity demand data, and generates, for another customer group, an electricity demand forecasting model that represents the relationship between the fluctuation factor data including other fluctuation factors not included in the fluctuation factor data for the one customer group and the electricity demand data.

An electric power demand forecasting system according to another aspect of the present disclosure includes a data storage unit that stores, for each of a plurality of customer groups of an electric power seller, training data including actual data on electric power demand fluctuation factors and actual data on electric power demand in a database; a model generation unit that generates, for each of a plurality of customer groups, a demand forecast model that represents the relationship between the fluctuation factor data and electric power demand data based on the plurality of training data stored in the database; an input information acquisition unit that acquires predicted data on the fluctuation factors for each of the plurality of customer groups; an individual demand forecasting unit that calculates predicted data on electric power demand based on the predicted data on the fluctuation factors and the demand forecasting model for each of the plurality of customer groups; and a total demand forecasting unit that calculates predicted data on total electric power demand for the electric power seller based on the predicted data on electric power demand calculated for each of the plurality of customer groups, and the plurality of customer groups are grouped according to differences in the fluctuation factors that affect electric power demand, and customers who are likely to have a similar relationship between the fluctuation factor data and electric power demand data belong to the same customer group.

An electric power demand forecasting method according to yet another aspect of the present disclosure includes: accumulating, for each of a plurality of customer groups of an electric power seller, training data including actual data on electric power demand fluctuation factors and actual data on electric power demand in a database; generating, for each of the plurality of customer groups, a demand forecasting model that represents a relationship between the fluctuation factor data and electric power demand data based on the plurality of training data accumulated in the database; acquiring forecast data of the fluctuation factors for each of the plurality of customer groups; calculating, for each of the plurality of customer groups, forecast data of electric power demand based on the forecast data of the fluctuation factors and the demand forecasting model; and calculating forecast data of total electric power demand for the electric power seller based on the forecast data of electric power demand calculated for each of the plurality of customer groups. Generating the demand forecasting model includes generating, for one customer group, an electric power demand forecasting model that represents a relationship between the fluctuation factor data including one fluctuation factor and the electric power demand data; and generating, for another customer group, an electric power demand forecasting model that represents a relationship between the fluctuation factor data including another fluctuation factor not included in the fluctuation factor data for the one customer group and the electric power demand data.

An electricity demand forecasting method according to yet another aspect of the present disclosure includes: accumulating, for each of a plurality of customer groups of an electricity seller, training data including actual data on electricity demand fluctuation factors and actual electricity demand data in a database; generating, for each of the plurality of customer groups, a demand forecasting model representing a relationship between the fluctuation factor data and electricity demand data based on the plurality of training data accumulated in the database; acquiring forecast data of the fluctuation factors for each of the plurality of customer groups; calculating, for each of the plurality of customer groups, forecast data of electricity demand based on the forecast data of the fluctuation factors and the demand forecasting model; and calculating forecast data of total electricity demand for the electricity seller based on the forecast data of electricity demand calculated for each of the plurality of customer groups, wherein the plurality of customer groups are grouped according to differences in the fluctuation factors that affect electricity demand, and customers who are likely to have a similar relationship between the fluctuation factor data and electricity demand data belong to the same customer group.

The present disclosure provides a system and method that are effective in improving the accuracy of electricity demand forecasts.

1 is a schematic diagram illustrating a configuration of an electric power supply and demand adjustment system. FIG. 1 is a block diagram illustrating a configuration of a power demand forecasting system. 13 is a table illustrating an example of a forecast result of individual demand. FIG. 13 is a block diagram showing a modified example of the power demand prediction system. FIG. 13 is a schematic diagram showing an example of a first evaluation screen. FIG. 13 is a schematic diagram showing an example of a second evaluation screen. 1 is a flowchart illustrating a procedure for generating a demand forecasting model. 1 is a flowchart illustrating a demand forecasting procedure. 11 is a flowchart illustrating a procedure for correcting a demand forecasting model.

The following describes the embodiments in detail with reference to the drawings. In the description, the same elements or elements having the same functions are given the same reference numerals, and duplicate descriptions are omitted.

[Electricity supply and demand adjustment system]
(Overall composition)
The electricity supply and demand adjustment system 1 shown in FIG. 1 is a system that adjusts the relationship between the demand and supply of electricity at a retail electricity supplier. The retail electricity business is a business that supplies electricity generated by a power generation facility owned by the retail electricity supplier, electricity purchased from a wholesale electricity exchange, etc., to customers (consumers). The retail electricity supplier supplies electricity to customers via the power grid of a power company. Specific examples of customers include customers who mainly consume electricity in offices and customers who mainly consume electricity in production factories. There are no particular restrictions on the facilities where the customers consume electricity, and the facilities are not limited to offices and production factories. For example, the facilities where the customers mainly consume electricity may be waste treatment facilities or stores.

To ensure the stability of the power grid, electricity retailers are required to "balance the supply and demand of electricity." "Balancing the supply and demand of electricity" refers to suppressing the discrepancy between the demand and supply of electricity in 30-minute increments. Under the "balancing the supply and demand of electricity" system, electricity retailers must submit a next-day supply and demand plan in 30-minute increments to the Organization for Cross-regional Coordination of Transmission Operators in advance. Electricity retailers subsequently settle an amount proportional to the difference (imbalance) between the submitted next-day supply and demand plan and the actual supply and demand. Electricity supply and demand adjustment system 1 is a system that supports electricity retailers under the "balancing the supply and demand of electricity" system.

As shown in FIG. 1, the power supply and demand adjustment system 1 includes a power demand forecasting system 10, a data collection system 20, and an administrator terminal 30. The power demand forecasting system 10, the data collection system 20, and the administrator terminal 30 are connected to each other via a communication network NW such as the Internet.

The electricity demand forecasting system 10 is a system that forecasts electricity demand for retail electricity suppliers (electricity sellers). For example, the electricity demand forecasting system 10 forecasts the electricity demand for the next day for retail electricity suppliers in 30-minute increments. Note that "30-minute increments" is merely an example, and there are no particular limitations on the time increments for which the electricity demand is forecast. For example, the electricity demand forecasting system 10 may forecast the electricity demand in 1-hour increments, 15-minute increments, etc.

The data collection system 20 is connected to the electricity meters 41 of multiple customers 40 of the retail electricity supplier via the communication network NW, and acquires actual data on electricity demand (e.g., actual data on electricity consumed) from each electricity meter 41. The data collection system 20 may also acquire the actual data on electricity demand from a server of the electric power company that owns the power grid.

The data collection system 20 also acquires performance data on variable factors for multiple customers 40. Variable factors include any factor that can cause fluctuations in electricity demand. Specific examples of variable factors include weather, day of the week, and whether it is a business day or a holiday. If the customer 40 is a production factory, the production status can also be a variable factor. The production status includes the type and number of products produced in a specified period of time. As an example, the data collection system 20 acquires the production status from the consumer through regular transmission.

Furthermore, the data collection system 20 also acquires actual data on the amount of power generated at the retail electricity supplier's own power generation equipment. For example, the data collection system 20 is connected to an electricity meter 51 of the retail electricity supplier's power generation equipment 50 via the communication network NW, and acquires actual data on the amount of power generated from the electricity meter 51. The data collection system 20 may also acquire actual data on electricity demand from a server of the electric power company that owns the power system.

The administrator terminal 30 is a terminal used by the administrator of the retail electricity supplier, and has a monitor (display) and an input device. Specific examples of the monitor include a liquid crystal monitor or an organic EL monitor. Specific examples of the input device include a keyboard or a mouse. The input device may be integrated with the monitor as a so-called touch panel.

(Electricity demand forecasting system)
The configuration of the power demand forecasting system 10 will be described in detail below. The power demand forecasting system 10 is configured to execute the following operations: for each of a plurality of customer groups of a retail electricity supplier, accumulate in a database training data including actual data on fluctuation factors of power demand and actual data on power demand, generate, for each of the plurality of customer groups, a demand forecasting model representing a relationship between the fluctuation factor data and power demand data based on the plurality of training data accumulated in the database, acquire prediction data of the fluctuation factors for each of the plurality of customer groups, calculate prediction data of power demand for each of the plurality of customer groups based on the prediction data of the fluctuation factors and the demand forecasting model, and calculate prediction data of a total power demand for a power seller based on the prediction data of power demand calculated for each of the plurality of customer groups.

The variable factors can vary greatly depending on the customer. For example, for customers who consume electricity in offices, weather and the like tend to be major factors, whereas for customers who consume electricity in production factories, production conditions tend to be major factors. In response to this, the present power demand forecasting system generates a power demand forecasting model for each of multiple customer groups, calculates forecast data for power demand for each of the multiple customer groups based on the power demand forecasting model, and calculates forecast data for total power demand for power sellers based on the forecast data for power demand calculated for each of the multiple customer groups. As a result, it is possible to group customers according to differences in factors that affect power demand, predict the power demand for each group with high accuracy, and aggregate the forecast results to predict total power demand with high accuracy. This is therefore effective in improving the accuracy of power demand forecasts.

As illustrated in FIG. 1, the power demand forecasting system 10 has a circuit 190. The circuit 190 has a processor 191, a memory 192, a storage 193, and a communication port 194. The storage 193 includes one or more non-volatile storage media (e.g., a hard disk or a flash memory) that can be read by a computer. The storage 193 stores a program for causing the power demand forecasting system 10 to execute the following operations: accumulating, for each of a plurality of customer groups of a retail electricity supplier, training data including actual data on fluctuation factors of power demand and actual data on power demand; generating, for each of a plurality of customer groups, a demand forecasting model that represents the relationship between the fluctuation factor data and the power demand data based on the plurality of training data accumulated in the database; acquiring forecast data of the fluctuation factors for each of the plurality of customer groups; calculating, for each of a plurality of customer groups, forecast data of power demand based on the forecast data of the fluctuation factors and the demand forecasting model; and calculating forecast data of the total power demand for the power selling company based on the forecast data of the power demand calculated for each of the plurality of customer groups. Storage 193 may include an area for storing the above programs as well as an area for storing the above database.

The memory 192 includes one or more volatile storage media (e.g., random access memory). The memory 192 temporarily stores the programs loaded from the storage 193 and the results of calculations by the processor 191. The processor 191 includes one or more calculation devices. The processor 191 executes the above programs in cooperation with the memory 192. The communication port 194 is connected to the data collection system 20 and the administrator terminal 30 via the communication network NW, and performs information communication between the data collection system 20 and the administrator terminal 30 in response to commands from the processor 191. Note that the connection here means that information communication is possible, and does not necessarily mean a physical connection by wire. The same applies below. In other words, at least a part of the communication network NW may be a wireless communication path.

In FIG. 1, an example is shown in which the power demand forecasting system 10 is configured with one set of computers, but the power demand forecasting system 10 may be configured with multiple sets of computers connected to each other via a communication network NW.

Figure 2 is a block diagram showing various functions realized by the processor 191 working in cooperation with the memory 192 to execute the above program, as block-shaped components (hereinafter referred to as "functional blocks"). In the example of Figure 2, the power demand forecasting system 10 has, as functional blocks, a performance information acquisition unit 111, an encoding processing unit 113, a data accumulation unit 114, a learning database 115, a model generation unit 116, a model holding unit 117, an input information acquisition unit 112, an individual demand forecasting unit 121, a total demand forecasting unit 122, and a forecast result output unit 123.

The performance information acquisition unit 111 acquires performance data of fluctuation factors of power demand and corresponding performance data of power demand from the data collection system 20 for each of the multiple customer groups. The fact that the performance data of power demand corresponds to the performance data of the fluctuation factors means that the target period of the performance data of power demand and the target period of the performance data of the fluctuation factors at least partially overlap. The multiple customer groups are groups that classify multiple customers of the retail electricity supplier. The multiple customers are grouped so that customers who are likely to have a similar relationship between the fluctuation factor data and the power demand data belong to the same customer group. Each of the multiple customer groups includes one or more customers of the retail electricity supplier. The performance information acquisition unit 111 acquires performance data indicating the actual value of the power consumption amount for each of 48 time frames obtained by dividing 24 hours into 30 minutes for each of the multiple customer groups.

The performance information acquisition unit 111 may acquire performance data of different variable factors for two or more customer groups among the multiple customer groups. For example, the multiple customer groups may include a first group that consumes electricity in an office and a second group that consumes electricity in a production factory.

In the first group, there is a tendency for the correlation between the variable factor data, which includes at least the weather, and the power demand data to be strong. Therefore, the performance information acquisition unit 111 may acquire performance data on variable factors, which includes at least the weather, for the first group. As an example, the performance information acquisition unit 111 acquires performance data on variable factors, which include the weather, the day of the week, and whether it is a business day or a holiday, for the first group.

In the second group, there is a tendency for the relationship between the variable factor data including at least the production status and the electricity demand data to be strong. Therefore, the performance information acquisition unit 111 may acquire performance data of variable factors including at least the production status for the second group. As an example, the performance information acquisition unit 111 acquires performance data of variable factors including weather and production status for the second group. The production status includes, for example, information representing the production object and the production quantity. In the second group, there may be a weak relationship between weather and electricity demand data. In this case, the variable factor for the second group does not need to include weather.

Customers who consume electricity in offices tend to have a similar relationship between the fluctuation factor data and the electricity demand data compared to customers who consume electricity in production plants. Therefore, the number of customers included in the first group may be greater than the number of customers included in the second group. The number of customers included in the second group may be 1.

The relationship between the variable factor data and the power demand data tends to be significantly different between customers who consume electricity in a production factory, depending on the difference in the production object. Therefore, the multiple customer groups may further include a third group that consumes electricity in a production factory where the production object is different from the production factory where the second group consumes electricity. The number of customers included in the third group may be one. The production objects being different from each other includes the production objects being the same type but having different specifications from each other, and the production objects being different types from each other. As an example, the performance information acquisition unit 111 also acquires performance data of the variable factors including the weather and the production status for the third group. The production status includes, for example, information representing the production object and the production quantity. In the third group, the relationship between the weather and the power demand data may be weak. In this case, the variable factor of the third group does not need to include the weather.

When the performance data of the variable factors includes a categorical variable, the encoding processing unit 113 converts the categorical variable into a quantitative variable by encoding. For example, the performance data of the production status may include a categorical variable representing the production status. A specific example of a categorical variable representing the production status is the name of a production pattern in which the production quantity for each type of production object is determined. In this case, the encoding processing unit 113 quantifies the production status by encoding. For example, the encoding processing unit 113 converts the categorical variable representing the production status into a quantitative variable by encoding.

The encoding process in this case may include a target encoding process in which the magnitude of power demand is used as the objective variable. For example, the encoding processing unit 113 converts a categorical variable representing the production status into a quantitative variable representing the magnitude of power demand based on a conversion table prepared in advance. The conversion table, for example, determines a numerical value representing the magnitude of power demand for each of a plurality of categorical variables. Hereinafter, the numerical value representing the magnitude of power demand is referred to as a "power demand value." For example, the encoding processing unit 113 extracts the power demand value associated with the categorical variable representing the production status from the conversion table as the value of the quantitative variable.

The encoding process performed by the encoding unit 113 is not necessarily limited to target encoding. The encoding unit 113 may convert categorical variables into quantitative variables by label encoding, or may convert categorical variables into quantitative variables by one-hot encoding.

The data accumulation unit 114 accumulates teacher data including actual data on fluctuation factors of electricity demand and actual data on electricity demand in the learning database 115 (database) for each of multiple customer groups. For example, the data accumulation unit 114 accumulates actual data on fluctuation factors of electricity demand acquired by the actual information acquisition unit 111 from the data collection system 20 and actual data on the corresponding electricity demand in the learning database 115 for each of multiple customer groups. When the actual data on fluctuation factors of electricity demand includes categorical variables, the data accumulation unit 114 may accumulate the actual data on the fluctuation factors after the categorical variables are converted into quantitative variables by the encoding processing unit 113 in the learning database 115.

The model generation unit 116 generates a demand forecasting model representing the relationship between the fluctuation factor data and the electricity demand data for each of multiple customer groups based on multiple training data stored in the learning database 115.

The model generation unit 116 may generate a demand forecasting model by at least the random forest method. In the random forest method, the model generation unit 116 randomly generates multiple decision trees based on performance data of variable factors included in the teacher data, and constructs each of the multiple decision trees based on the teacher data. For example, the model generation unit 116 constructs the decision tree to be constructed as follows. That is, the model generation unit 116 randomly samples the teacher data, provides the sampled data to the top node of the decision tree, and obtains a branch function for the node. Next, the model generation unit 116 divides the teacher data into two systems based on the branch function, provides the two nodes with the node as a parent, and obtains the branch function for the two nodes. Thereafter, the model generation unit 116 continues the above process for lower nodes until a preset termination condition is met, and completes the construction of the decision tree.

The model generation unit 116 may generate a demand forecasting model by ensemble learning that combines a random forest method with one or more other learning methods. The one or more other learning methods may include at least a recurrent neural network method. A recurrent neural network is a neural network that has a recurrent structure that loops information stored in the past in the intermediate layer. The recurrent structure gives the neural network memory, improving the prediction accuracy for time series data.

The model generation unit 116 may generate a first demand forecast model for the first group that represents the relationship between the variable factor data including at least the weather and the electricity demand data, generate a second demand forecast model for the second group that represents the relationship between the variable factor data including at least the production status and the electricity demand data, and generate a third demand forecast model for the third group that represents the relationship between the variable factor data including at least the production status and the electricity demand data.

As an example, the model generation unit 116 generates a first demand forecast model that represents the relationship between the electricity demand data and the variable factor data, which includes the weather, the day of the week, and whether it is a business day or a holiday. The model generation unit 116 generates a second demand forecast model and a third demand forecast model that represent the relationship between the electricity demand data and the variable factor data, which includes the weather and production status.

As described above, the data accumulation unit 114 may include the results of the categorical variables converted into quantitative variables by the encoding processing unit 113 in the performance data of the variable factors. In this case, the variable factor data including the production status may include data in which the production status has been quantified by encoding processing.

The model generation unit 116 stores the demand forecasting model 118 generated for each of the multiple customer groups in the model storage unit 117.

The input information acquisition unit 112 acquires forecast data of fluctuation factors of electricity demand for each of multiple customer groups. For example, the input information acquisition unit 112 acquires forecast data of fluctuation factors of electricity demand for each of multiple customer groups from the data collection system 20. For example, the input information acquisition unit 112 acquires forecast data of the above fluctuation factors for the next day from the data collection system 20.

The attributes of the prediction data of the variable factors acquired by the input information acquisition unit 112 for each of the multiple customer groups are equal to the attributes of the actual data of the variable factors acquired by the actual data acquisition unit 111 for each of the multiple customer groups. For example, the input information acquisition unit 112 acquires prediction data of the variable factors including the weather, day of the week, and whether it is a business day or a holiday for the first group, and acquires prediction data of the variable factors including the weather and production status for the second and third groups. The prediction data of the production status corresponds to a production plan or a factory operation plan. Even if the prediction data of the variable factors includes categorical variables, the encoding processing unit 113 converts the categorical variables into quantitative variables by encoding processing.

The individual demand forecasting unit 121 calculates, for each of multiple customer groups, forecast data for electricity demand based on the forecast data for fluctuation factors acquired by the input information acquisition unit 112 and the demand forecast model 118 stored in the model holding unit 117. When the forecast data for fluctuation factors for electricity demand includes categorical variables, the individual demand forecasting unit 121 may calculate the forecast data for electricity demand based on the forecast data for fluctuation factors after the categorical variables are converted into quantitative variables by the encoding processing unit 113 and the demand forecasting model 118 stored in the model holding unit 117.

The format of the electricity demand forecast data generated by the individual demand forecasting unit 121 for each of the multiple customer groups may be the same as the format of the electricity demand historical data acquired by the historical information acquisition unit 111 for each of the multiple customer groups. For example, as illustrated in FIG. 3, the individual demand forecasting unit 121 calculates, for each of the multiple customer groups, forecast data indicating the predicted value of the amount of electricity consumed for each of 48 time frames obtained by dividing 24 hours into 30-minute intervals. In the illustration, each horizontal line represents the electricity demand forecast data for one customer group.

The total demand forecasting unit 122 calculates forecast data of total electricity demand for the retail electricity supplier based on the forecast data of electricity demand calculated for each of the multiple customer groups. For example, the total demand forecasting unit 122 calculates forecast data of total electricity demand indicating the total electricity consumption for each of the 48 time frames by adding up the forecast values of the electricity consumption for each of the multiple customer groups for each of the above-mentioned 48 time frames.

The prediction result output unit 123 outputs the prediction data of the total power demand calculated by the total demand prediction unit 122. Outputting the prediction data includes displaying the prediction data. For example, the prediction result output unit 123 displays the prediction data on the monitor of the administrator terminal 30.

The electricity demand forecasting system 10 may be further configured to evaluate an error between the electricity demand forecast data and the electricity demand actual data for each of the multiple customer groups. In this case, the electricity demand forecasting system 10 may be further configured to revise new electricity demand forecast data for each of the multiple customer groups based on the error evaluation results.

The electricity demand forecasting system 10 may be further configured to correct the demand forecasting model for each of a plurality of customer groups based on the error evaluation results. For example, as shown in FIG. 4, the electricity demand forecasting system 10 further includes, as functional blocks, an error evaluation unit 131, an evaluation image display unit 132, a prediction correction unit 133, and a model correction unit 134.

The error evaluation unit 131 evaluates the error between the forecast data of the power demand and the actual data of the power demand for each of the multiple customer groups. For example, the error evaluation unit 131 calculates the difference between the forecast value and the actual value for each unit time (for example, 30 minutes) and sums up the calculation results of the difference for one day. There is no particular limit to the period to be summed, and it may be shorter or longer than one day. For example, the error evaluation unit 131 calculates the difference between the forecast value and the actual value of the power consumption for each of the above 48 frames of time, sums up the calculation results of the difference for each of the 48 frames of time, and calculates the sum as the evaluation value of the error. According to this calculation method, the positive difference and the negative difference cancel each other out. Therefore, if the positive and negative of the difference randomly changes for each time frame, the absolute value of the evaluation value becomes small. On the other hand, if the positive and negative of the difference tend to match between time frames, the absolute value of the evaluation value tends to become large. If the prediction accuracy of the demand forecast model itself decreases, it is considered that the positive and negative of the difference will easily match between time frames. Therefore, the above calculation method makes it possible to calculate an evaluation value that is likely to represent a decline in the prediction accuracy of the demand forecasting model itself.

The evaluation image display unit 132 displays on a monitor an evaluation image that visualizes the evaluation results of the error for each of the multiple customer groups. For example, the evaluation image display unit 132 displays the evaluation image on the monitor of the administrator terminal 30. For example, the evaluation image display unit 132 may display on a monitor a first evaluation image that visualizes the change over time in the evaluation results of the error for each of the multiple customer groups.

As an example, the evaluation image display unit 132 may display on the monitor a first evaluation image in which colors representing the evaluation results of the error are mapped according to multiple customer groups and the evaluation time. For example, as shown in FIG. 5, the evaluation image display unit 132 displays on the monitor a first evaluation image 140 in which cells 141 indicating the above-mentioned evaluation value by at least one of the three color attributes (hue, brightness, saturation) are mapped according to multiple customer groups and the evaluation time. Each cell 141 may represent the error evaluation result by hatching or dots, etc. Note that the average color of the cell 141 may change depending on the pattern shape, thickness, size, or density of the hatching or dots. Therefore, representing the error evaluation result by hatching or dots, etc., may be an example of representing the error evaluation result by color.

In the example of Figure 5, a horizontal row shows the evaluation value for one customer group on a daily basis. The multiple cells 141 in this row are arranged so that the more to the right the cells 141 have newer evaluation dates. This arrangement makes it possible to visualize the daily changes in evaluation values (an example of changes over time). A vertical row shows the daily evaluation values for multiple customer groups.

In the example of FIG. 5, the area is divided into a latest data display area 143, which contains evaluation values for the week including the latest date, and a past data display area 142, which contains evaluation values prior to that. In the latest date column, no cell 141 is placed in a location where an evaluation value has not yet been calculated.

The evaluation image display unit 132 may display on the monitor a second evaluation image that visualizes the relationship between the error evaluation result and the correction criterion of the demand forecasting model for each of the multiple customer groups. The evaluation image display unit 132 may display on the monitor an evaluation image that includes a bar graph representing the error evaluation result and a correction criterion line representing the correction criterion of the demand forecasting model for each of the multiple customer groups.

For example, as shown in FIG. 6, the evaluation image display unit 132 may display on the monitor a second evaluation image 150 including a bar graph 151 representing the absolute value of the above-mentioned evaluation value and a correction reference line 152 representing the correction criterion of the demand forecast model for each of the multiple customer groups. According to FIG. 6, the need for correction of the demand forecast model is clearly indicated for each of the multiple customer groups depending on whether the bar graph 151 reaches the correction reference line 152. As an example, the evaluation image display unit 132 displays both the first evaluation image 140 and the second evaluation image 150 on the monitor of the administrator terminal 30. The evaluation image display unit 132 may change the color of each bar graph 151 based on the relationship between the correction criterion and the evaluation result of the error. For example, a bar graph that exceeds the correction criterion (correction reference line 152) may be displayed in a first color (e.g., red), a bar graph that is likely to exceed the correction criterion (within a predetermined range from the correction criterion) may be displayed in a second color (e.g., orange), and a bar graph that falls below the correction criterion by a predetermined value or more may be displayed in a third color (e.g., blue). Also, as shown in FIG. 6, the color of each bar graph may be changed for each part based on its relationship with the correction criterion. For example, the part of the bar graph that exceeds the correction criterion may be shown in a first color, the part that is likely to exceed the correction criterion may be shown in a second color, and the part that falls below the correction criterion by a predetermined level or more may be shown in a third color. A gradation may be applied between the colors.

The prediction correction unit 133 corrects new forecast data of power demand for each of the multiple customer groups based on the error evaluation result. The prediction correction unit 133 may correct the new forecast data of power demand for each of the multiple customer groups based on user input after the first evaluation image 140 is displayed. For example, the prediction correction unit 133 may display the first evaluation image 140 on the monitor of the administrator terminal 30, and then correct the new forecast data of power demand based on input from an input device of the administrator terminal 30. The prediction correction unit 133 may automatically correct the new forecast data of power demand based on the error evaluation result. A specific example of automatic correction is adding or subtracting the error evaluation result to or from the new forecast data of power demand.

The first evaluation image 140 indicates whether the tendency for evaluation values to be positive or negative continues over multiple days. If the tendency for evaluation values to be positive or negative continues over multiple days, it is highly likely that the same trend will also appear in the new predicted data. Therefore, by correcting the new predicted data so as to counteract the tendency for evaluation values to be positive or negative, it is possible to improve the accuracy of the predicted data.

The model correction unit 134 corrects the demand forecast model for each of the multiple customer groups based on the error evaluation results. The model correction unit 134 may correct the demand forecast model for each of the multiple customer groups based on user input after the second evaluation image 150 is displayed. For example, the model correction unit 134 may display the second evaluation image 150 on the monitor of the administrator terminal 30, and then correct the demand forecast model based on input via an input device of the administrator terminal 30.

As an example, when an instruction to modify the demand forecasting model is input by an input device of the administrator terminal 30, the model correction unit 134 requests the model generation unit 116 to regenerate the demand forecasting model. When regeneration of the demand forecasting model is requested by the model correction unit 134, the model generation unit 116 regenerates the demand forecasting model based on the teacher data accumulated in the learning database 115 at that time, and stores the regenerated demand forecasting model in the model holding unit 117. The model correction unit 134 may automatically correct the demand forecasting model based on the error evaluation result. For example, the model correction unit 134 may automatically request the model generation unit 116 to regenerate the demand forecasting model based on the error evaluation result.

[Electricity demand forecast procedure]
As an example of an electricity demand forecasting method, an electricity demand forecasting procedure executed by the electricity demand forecasting system 10 is illustrated. This procedure includes accumulating, for each of a plurality of customer groups of the retail electricity supplier, training data including actual data on fluctuation factors of electricity demand and actual data on electricity demand in the learning database 115, generating, for each of the plurality of customer groups, a demand forecasting model 118 representing a relationship between the fluctuation factor data and electricity demand data based on the plurality of training data accumulated in the learning database 115, acquiring forecast data of the fluctuation factors for each of the plurality of customer groups, calculating forecast data of electricity demand for each of the plurality of customer groups based on the forecast data of the fluctuation factors and the demand forecasting model 118, and calculating forecast data of total electricity demand for the retail electricity supplier based on the forecast data of electricity demand calculated for each of the plurality of customer groups. Below, this procedure is illustrated in detail by dividing it into a demand forecasting model generating procedure, a power demand forecasting data calculating procedure, and a demand forecasting model regenerating procedure.

(Demand forecast model generation procedure)
7, the power demand forecasting system 10 first executes steps S01 and S02. In step S01, the performance information acquisition unit 111 acquires performance data on fluctuation factors of power demand and corresponding performance data on power demand for each of a plurality of customer groups from the data collection system 20. In step S02, the encoding processing unit 113 checks whether the performance data on the fluctuation factors acquired by the performance information acquisition unit 111 includes a categorical variable.

If it is determined in step S02 that the performance data of the variation factors includes categorical variables, the power demand forecasting system 10 executes step S03. In step S03, the encoding processing unit 113 converts the categorical variables included in the performance data of the variation factors into quantitative variables through encoding processing.

Next, the power demand forecasting system 10 executes steps S04 and S05. If it is determined in step S02 that the performance data of the fluctuation factors does not include a categorical variable, the power demand forecasting system 10 executes steps S04 and S05 without executing step S03. In step S04, the data accumulation unit 114 accumulates the performance data of the fluctuation factors of the power demand acquired by the performance information acquisition unit 111 from the data collection system 20 and the corresponding performance data of the power demand in the learning database 115 for each of multiple customer groups. If the performance data of the fluctuation factors of the power demand includes a categorical variable, the data accumulation unit 114 accumulates the performance data of the fluctuation factors after the categorical variable is converted into a quantitative variable by the encoding processing unit 113 in the learning database 115. In step S05, the model generation unit 116 checks whether a learning command has been input by an input device or the like of the administrator terminal 30.

If it is determined in step S05 that a learning command has not been input, the power demand forecasting system 10 returns the process to step S01. Thereafter, until a learning command is input, the accumulation of training data for each of the multiple customer groups is repeated at a predetermined cycle.

If it is determined in step S05 that a learning command has been input, the electricity demand forecasting system 10 executes steps S06, S07, S08, and S09. In step S06, the model generation unit 116 selects one of the multiple customer groups as a target for generating a demand forecasting model. Hereinafter, in the explanation of steps S07 to S11, the customer group selected by the model generation unit 116 as a target for generating a demand forecasting model will be referred to as the "selected group."

In step S07, the model generation unit 116 extracts teacher data for the selected group from the teacher data stored in the learning database 115. In step S08, the model generation unit 116 generates a demand forecasting model 118 corresponding to the selected group by machine learning based on the extracted teacher data, and stores it in the model holding unit 117. In step S09, the model generation unit 116 checks whether the generation of demand forecasting models for all customer groups has been completed.

If it is determined in step S09 that there are still customer groups for which a demand forecast model has not been generated, the electricity demand forecasting system 10 executes step S11. In step S11, the model generation unit 116 selects the next customer group from one or more customer groups for which a demand forecast model has not been generated among the multiple customer groups. The electricity demand forecasting system 10 then returns the process to step S07. Thereafter, the selection of a customer group and the generation of a demand forecast model for the selected customer group are repeated until the generation of demand forecast models for all customer groups is completed. If it is determined in step S09 that the generation of demand forecast models for all customer groups is completed, the procedure for generating the demand forecast model is completed.

(Calculation procedure for forecast data of electricity demand)
As shown in Fig. 8, the power demand forecasting system 10 first executes steps S21, S22, and S23. In step S21, the input information acquisition unit 112 acquires forecast data of fluctuation factors of power demand for each of a plurality of customer groups. In step S22, the individual demand forecasting unit 121 selects one of the plurality of customer groups as a target for calculating forecast data. Hereinafter, the customer group selected by the individual demand forecasting unit 121 as a target for calculating forecast data is referred to as the "selected group." In step S23, the encoding processing unit 113 checks whether or not the forecast data of fluctuation factors for the selected group, among the forecast data of fluctuation factors acquired by the individual demand forecasting unit 121, includes a categorical variable.

If it is determined in step S23 that the prediction data of the variation factors for the selected group includes a categorical variable, the power demand prediction system 10 executes step S24. In step S24, the encoding processing unit 113 converts the categorical variable included in the prediction data of the variation factors for the selected group into a quantitative variable through encoding processing.

Next, the power demand forecasting system 10 executes steps S25 and S26. If it is determined in step S23 that the forecast data of the fluctuation factors for the selected group does not include a categorical variable, the power demand forecasting system 10 executes step S25 without executing step S24. In step S25, the individual demand forecasting unit 121 calculates forecast data of power demand based on the forecast data of the fluctuation factors for the selected group and the demand forecast model 118 of the selected group stored in the model holding unit 117. If the forecast data of the fluctuation factors includes a categorical variable, the individual demand forecasting unit 121 calculates forecast data of power demand based on the forecast data of the fluctuation factors after the categorical variable is converted into a quantitative variable by the encoding processing unit 113 and the demand forecasting model 118. In step S26, the individual demand forecasting unit 121 checks whether the calculation of the forecast data of power demand for all customer groups has been completed.

If it is determined in step S26 that there are still customer groups for which forecast data for electricity demand has not been calculated, the electricity demand forecasting system 10 executes step S27. In step S27, the individual demand forecasting unit 121 selects the next customer group from one or more customer groups for which forecast data for electricity demand has not been calculated among the multiple customer groups. The electricity demand forecasting system 10 then returns the process to step S23. Thereafter, the selection of a customer group and the calculation of forecast data for electricity demand for the selected customer group are repeated until the calculation of forecast data for electricity demand for all customer groups is completed.

If it is determined in step S26 that calculation of the electricity demand forecast data for all customer groups has been completed, the electricity demand forecasting system 10 executes step S31. In step S31, the forecast correction unit 133 checks whether a correction command for the electricity demand forecast data for any of the multiple customer groups has been input via the input device of the administrator terminal 30.

If it is determined in step S31 that a correction command for the electricity demand forecast data has been input for any of the multiple customer groups, the electricity demand forecast system 10 executes step S32. In step S32, the forecast correction unit 133 corrects new electricity demand forecast data based on the correction command input via the input device of the administrator terminal 30.

Next, the electricity demand forecasting system 10 executes step S33. If it is determined in step S31 that a command to correct the electricity demand forecast data has not been input for any customer group, the electricity demand forecasting system 10 executes step S33 without executing step S32. In step S33, the total demand forecasting unit 122 checks whether or not a command to output the total electricity demand forecast data has been input via the input device of the administrator terminal 30.

If it is determined in step S33 that a command to output forecast data for total power demand has not been input, the power demand forecast system 10 returns the process to step S31. Thereafter, until a command to output forecast data for total power demand is input, the system repeatedly modifies new forecast data for power demand in response to input of a correction instruction. Note that, since the error evaluation result is not visualized until the first evaluation image 140 is displayed in step S37 described below, step S32 is not executed, at least in the first cycle.

If it is determined in step S33 that a command to output forecast data for total electricity demand has been input, the electricity demand forecasting system 10 executes steps S34, S35, S36, and S37. In step S34, the total demand forecasting unit 122 calculates forecast data for total electricity demand for the retail electricity supplier based on the forecast data for electricity demand calculated for each of multiple customer groups. In step S35, the forecast result output unit 123 outputs the forecast data for total electricity demand calculated by the total demand forecasting unit 122 to the data collection system 20.

In step S36, the performance information acquisition unit 111 acquires performance data on the fluctuation factors of electricity demand and corresponding performance data on electricity demand for each of multiple customer groups from the data collection system 20. In step S37, the encoding processing unit 113 checks whether the performance data on the fluctuation factors acquired by the performance information acquisition unit 111 includes a categorical variable.

If it is determined in step S37 that the performance data of the variation factors includes categorical variables, the power demand forecasting system 10 executes step S38. In step S38, the encoding processing unit 113 converts the categorical variables included in the performance data of the variation factors into quantitative variables through encoding processing.

Next, the power demand forecasting system 10 executes steps S39, S41, and S42. If it is determined in step S37 that the performance data of the fluctuation factors does not include a categorical variable, the power demand forecasting system 10 executes steps S39, S41, and S42 without executing step S38. In step S39, the data accumulation unit 114 accumulates the performance data of the fluctuation factors of power demand acquired by the performance information acquisition unit 111 from the data collection system 20 and the corresponding performance data of power demand in the learning database 115 for each of multiple customer groups. If the performance data of the fluctuation factors of power demand includes a categorical variable, the data accumulation unit 114 accumulates the performance data of the fluctuation factors after the categorical variable is converted into a quantitative variable by the encoding processing unit 113 in the learning database 115.

In step S41, the error evaluation unit 131 evaluates the error between the forecasted electricity demand data and the actual electricity demand data for each of the multiple customer groups. In step S42, the evaluation image display unit 132 displays an evaluation image that visualizes the error evaluation results for each of the multiple customer groups on a monitor. For example, the evaluation image display unit 132 displays a first evaluation image 140 and a second evaluation image 150 on the monitor of the administrator terminal 30.

Then, the electricity demand forecasting system 10 returns the process to step S21. From then on, the first evaluation image 140 and the second evaluation image 150 are displayed on the monitor of the administrator terminal 30, and the error evaluation result is visualized. This makes it possible to input a correction command for the electricity demand forecast data, for example, based on the first evaluation image 140, via the input device of the administrator terminal 30. Therefore, step S32 can be executed in response to the input of the correction command.

(Procedure for regenerating demand forecast model)
As shown in Fig. 9, the power demand forecasting system 10 executes steps S51, S52, S53, and S54 in order. In step S51, the model correction unit 134 waits for a command to correct the demand forecasting model to be input by the input device of the manager terminal 30 based on the second evaluation image 150. In step S52, the model correction unit 134 requests the model generation unit 116 to regenerate a demand forecasting model for the customer group for which a command to correct the demand forecasting model has been input. Hereinafter, the customer group for which a command to correct the demand forecasting model has been input is referred to as the "group to be corrected."

In step S53, the model generation unit 116 extracts the teacher data for the group to be corrected from the teacher data stored in the learning database 115. In step S54, the model generation unit 116 regenerates the demand forecast model 118 corresponding to the group to be corrected by machine learning based on the extracted teacher data, and stores it in the model holding unit 117. Thereafter, the electricity demand forecasting system 10 returns the process to step S51.

[Effects of this embodiment]
As described above, the electricity demand forecasting system 10 includes a data storage unit 114 that accumulates teacher data including actual data on electricity demand fluctuation factors and actual data on electricity demand in a learning database 115 for each of the electricity seller's multiple customer groups, a model generation unit 116 that generates a demand forecast model 118 that represents the relationship between the fluctuation factor data and electricity demand data based on the multiple teacher data accumulated in the learning database 115 for each of the multiple customer groups, an input information acquisition unit 112 that acquires predicted data of fluctuation factors for each of the multiple customer groups, an individual demand forecasting unit 121 that calculates predicted data of electricity demand for each of the multiple customer groups based on the predicted data of fluctuation factors and the demand forecasting model 118, and a total demand forecasting unit 122 that calculates predicted data of total electricity demand for the electricity seller based on the predicted data of electricity demand calculated for each of the multiple customer groups.

Factors that affect power demand can vary greatly depending on the customer. For example, for customers who consume power in offices, weather and the like tend to be major factors, whereas for customers who consume power in production plants, production conditions tend to be major factors. In response to this, the power demand forecasting system 10 generates a demand forecasting model 118 for each of multiple customer groups, calculates forecast data for power demand for each of the multiple customer groups based on the demand forecasting model 118, and calculates forecast data for total power demand for the power seller based on the forecast data for power demand calculated for each of the multiple customer groups. For this reason, it is possible to group customers according to differences in factors that affect power demand, predict the power demand for each group with high accuracy, and aggregate the forecast results to predict the total power demand with high accuracy. This is therefore effective in improving the accuracy of power demand forecasts.

The model generation unit 116 may generate a demand forecasting model using at least the random forest method. In this case, a demand forecasting model with high prediction accuracy can be easily generated.

The model generation unit 116 may generate a demand forecasting model by ensemble learning that combines the random forest method with one or more other learning methods. In this case, a demand forecasting model with higher prediction accuracy can be generated.

The one or more other learning techniques may include at least a recurrent neural network method. In this case, a demand forecasting model with higher forecasting accuracy can be generated.

The multiple customer groups may include a first group that consumes electricity in an office and a second group that consumes electricity in a production plant, and the model generation unit 116 may generate a first demand forecast model for the first group that represents the relationship between the variable factor data, including at least the weather, and the electricity demand data, and generate a second demand forecast model for the second group that represents the relationship between the variable factor data, including at least the production status, and the electricity demand data. In this case, customers are divided into groups according to differences in factors that affect electricity demand, and the electricity demand for each group is predicted with high accuracy, and the prediction results are compiled to predict the total electricity demand with high accuracy.

The number of customers included in the first group may be greater than the number of customers included in the second group. There is a warning that the fluctuation factors of electricity demand tend to differ between customers who consume electricity in production factories. On the other hand, the fluctuation factors of electricity demand tend to be similar between customers who consume electricity in offices. Therefore, by grouping more customers into the first group than the number of customers included in the second group, the reliability of the training data can be improved and the prediction accuracy of the first demand forecasting model can be improved.

The multiple customer groups may further include a third group that consumes electricity in a production factory that produces different products from those produced by the second group, and the model generation unit 116 may generate a third demand forecasting model for the third group that represents the relationship between the variable factor data, including at least the production status, and the electricity demand data. In this case, by grouping customers that produce different products, the forecasting accuracy of the demand forecasting model 118 for each group can be improved.

The variable factor data including the production status may include data in which the production status is quantified by encoding processing. In this case, it is even easier to generate a demand forecast model 118 for a group including customers who consume electricity in a production plant.

The encoding process may include a target encoding process in which the magnitude of electricity demand is used as the objective variable. In this case, the prediction accuracy of the electricity demand forecasting model for a group including customers who consume electricity in a production plant can be further improved.

The power demand forecasting system 10 may further include an error evaluation unit 131 that evaluates the error between the power demand forecast data and the power demand actual data for each of the multiple customer groups, and a forecast correction unit 133 that corrects new power demand forecast data based on the error evaluation results for each of the multiple customer groups. While it is difficult to obtain reproducibility in the error between the forecast data and the actual data for the total power demand, it is easy to obtain reproducibility in the error between the forecast data and the actual data for each customer group. Therefore, by performing error evaluation and correction of the power demand forecast data based on the error evaluation results for each of the multiple customer groups, it is possible to further improve the accuracy of the power demand forecast.

The electricity demand forecasting system 10 further includes an evaluation image display unit 132 that displays an evaluation image that visualizes the error evaluation results on a monitor for each of the multiple customer groups, and the forecast correction unit 133 may correct new forecast data for electricity demand for each of the multiple customer groups based on user input after the evaluation image is displayed. In this case, by visualizing the error evaluation results for each of the multiple customer groups and leaving it up to the user to correct the forecast data for electricity demand for each of the multiple customer groups, it is possible to effectively utilize the user's knowledge to generate forecast data for electricity demand.

The evaluation image display unit 132 may display on the monitor a first evaluation image that visualizes the change over time in the evaluation results of the error for each of multiple customer groups. In this case, the recurrence tendency of the error is shown for each of multiple customer groups, and the user can be prompted to make further appropriate corrections to the electricity demand forecast data.

The evaluation image display unit 132 may display on the monitor a first evaluation image 140 in which colors representing the evaluation results of the error are mapped according to multiple customer groups and the time of evaluation. In this case, the tendency for the error to reoccur can be more clearly shown for each of multiple customer groups.

The power demand forecasting system 10 may further include an error evaluation unit 131 that evaluates the error between the power demand forecast data and the power demand actual data for each of multiple customer groups, and a model correction unit 134 that corrects the demand forecasting model 118 based on the error evaluation results for each of multiple customer groups. The prediction accuracy of the demand forecasting model 118 may decrease due to changes in the customer's power facilities. For example, the prediction accuracy of the demand forecasting model 118 may decrease due to the customer's expansion of private power generation facilities, expansion of machine tools, and an increase in the proportion of unmanned processes. By evaluating the error and correcting the demand forecasting model based on the error evaluation results for each of multiple customer groups, the demand forecasting model can be corrected more accurately at a more appropriate time. This makes it easy to maintain high prediction accuracy of power demand.

The electricity demand forecasting system 10 further includes an evaluation image display unit 132 that displays an evaluation image that visualizes the error evaluation results on a monitor for each of the multiple customer groups, and the model correction unit 134 may correct the demand forecasting model 118 for each of the multiple customer groups based on a user input after the evaluation image is displayed. In this case, by visualizing the error evaluation results for each of the multiple customer groups and leaving it up to the user to correct the demand forecasting model 118 for each of the multiple customer groups, the user's knowledge can be effectively utilized to more appropriately correct the demand forecasting model 118.

The evaluation image display unit 132 may display on the monitor a second evaluation image that visualizes the relationship between the error evaluation results and the correction criteria for the demand forecasting model 118 for each of the multiple customer groups. In this case, the criteria for determining whether or not to correct the demand forecasting model 118 can be presented in an easy-to-understand manner for each of the multiple customer groups.

The evaluation image display unit 132 may display on the monitor a second evaluation image 150 including a bar graph 151 showing the error evaluation result and a correction reference line 152 showing the correction reference for the demand forecasting model 118 for each of the multiple customer groups. In this case, the criteria for determining whether or not to correct the demand forecasting model 118 can be presented more clearly for each of the multiple customer groups.

Although the embodiments have been described above, the present disclosure is not necessarily limited to the above-described embodiments, and various modifications are possible without departing from the spirit of the present disclosure.

10...electricity demand forecasting system, 112...input information acquisition unit, 114...data accumulation unit, 115...learning database (database), 116...model generation unit, 118...demand forecasting model, 121...individual demand forecasting unit, 122...total demand forecasting unit, 131...error evaluation unit, 132...evaluation image display unit, 133...forecast correction unit, 134...model correction unit, 140...first evaluation image (evaluation image), 150...second evaluation image (evaluation image), 151...bar graph, 152...corrected reference line.

Claims (22)

a data storage unit that stores, in a database, teacher data including actual data on fluctuation factors of power demand and actual data on power demand for each of a plurality of customer groups of the power selling company;
a model generation unit that generates, for each of the plurality of customer groups, a demand forecast model that represents a relationship between fluctuation factor data and electricity demand data based on a plurality of teacher data stored in the database;
an input information acquisition unit that acquires prediction data of the fluctuation factors for each of the plurality of customer groups;
an individual demand forecasting unit that calculates forecast data of electricity demand for each of the plurality of customer groups based on the forecast data of the fluctuation factors and the demand forecast model;
a total demand forecasting unit that calculates forecast data of a total electricity demand for the electricity seller based on the forecast data of the electricity demand calculated for each of the plurality of customer groups,
The model generation unit
generating an electricity demand forecasting model representing a relationship between fluctuation factor data including one fluctuation factor and electricity demand data for one customer group;
An electricity demand forecasting system that generates an electricity demand forecasting model for other customer groups, the electricity demand forecasting model representing the relationship between variable factor data including other variable factors not included in the variable factor data for the one customer group and electricity demand data. a data storage unit that stores, in a database, teacher data including actual data on fluctuation factors of power demand and actual data on power demand for each of a plurality of customer groups of the power selling company;
a model generation unit that generates, for each of the plurality of customer groups, a demand forecast model that represents a relationship between fluctuation factor data and electricity demand data based on a plurality of teacher data stored in the database;
an input information acquisition unit that acquires prediction data of the fluctuation factors for each of the plurality of customer groups;
an individual demand forecasting unit that calculates forecast data of electricity demand for each of the plurality of customer groups based on the forecast data of the fluctuation factors and the demand forecast model;
a total demand forecasting unit that calculates forecast data of a total electricity demand for the electricity seller based on the forecast data of the electricity demand calculated for each of the plurality of customer groups,
The plurality of customer groups are grouped according to differences in the variable factors affecting the electricity demand;
An electricity demand forecasting system, wherein customers who are likely to have a similar relationship between the fluctuation factor data and the electricity demand data belong to the same customer group. The electricity demand forecasting system according to claim 1 or 2, wherein the model generation unit generates the demand forecasting model using at least a random forest method. The power demand forecasting system of claim 3, wherein the model generation unit generates the demand forecasting model by ensemble learning that combines the random forest method with one or more other learning methods. The power demand forecasting system of claim 4, wherein the one or more other learning methods include at least a recurrent neural network method. the plurality of customer groups include a first group that consumes electricity in an office and a second group that consumes electricity in a production factory;
The model generation unit
generating a first demand forecasting model representing a relationship between variable factor data including at least weather and power demand data for the first group;
The power demand forecasting system according to any one of claims 1 to 5, further comprising: generating a second demand forecasting model for the second group that represents a relationship between variable factor data including at least production status and power demand data. The electricity demand forecasting system of claim 6, wherein the number of customers included in the first group is greater than the number of customers included in the second group. the plurality of customer groups further includes a third group that consumes electricity in a production factory that produces products different from those produced in the production factory in which the second group consumes electricity;
The power demand forecasting system according to claim 6 or 7, wherein the model generation unit generates a third demand forecasting model representing a relationship between variable factor data including at least a production status and power demand data for the third group. The power demand forecasting system according to any one of claims 6 to 8, wherein the fluctuation factor data including the production status includes data in which the production status is quantified by encoding processing. The power demand forecasting system according to claim 9, wherein the encoding process includes a target encoding process in which the magnitude of power demand is used as a target variable. an error evaluation unit that evaluates an error between the power demand forecast data and the power demand actual data for each of the plurality of customer groups;
The power demand forecasting system according to any one of claims 1 to 10, further comprising: a forecast correction unit that corrects new forecast data of power demand for each of the plurality of customer groups based on the error evaluation result. The electricity demand forecasting system according to claim 11, further comprising an evaluation image display unit that displays, on a monitor, an evaluation image that visualizes the evaluation result of the error for each of the plurality of customer groups. The electricity demand forecasting system according to claim 12, wherein the forecast correction unit corrects the new forecast data of the electricity demand for each of the multiple customer groups based on a user input after the evaluation image is displayed. The electricity demand forecasting system according to claim 13, wherein the evaluation image display unit displays on the monitor the evaluation image that visualizes the change over time in the evaluation result of the error for each of the multiple customer groups. The power demand forecasting system according to claim 14, wherein the evaluation image display unit displays on the monitor the evaluation image in which the colors representing the evaluation results of the errors are mapped according to the multiple customer groups and the evaluation time. an error evaluation unit that evaluates an error between the power demand forecast data and the power demand actual data for each of the plurality of customer groups;
The power demand forecasting system according to any one of claims 1 to 10, further comprising: a model correction unit that corrects the demand forecasting model based on the error evaluation result for each of the plurality of customer groups. The electricity demand forecasting system according to claim 16, further comprising an evaluation image display unit that displays, on a monitor, an evaluation image that visualizes the evaluation result of the error for each of the plurality of customer groups. The electricity demand forecasting system according to claim 17, wherein the model correction unit corrects the demand forecasting model for each of the plurality of customer groups based on a user input after the evaluation image is displayed. The electricity demand forecasting system according to claim 18, wherein the evaluation image display unit displays on the monitor the evaluation image that visualizes the relationship between the error evaluation result and the correction criterion of the demand forecasting model for each of the multiple customer groups. The electricity demand forecasting system of claim 19, wherein the evaluation image display unit displays on the monitor the evaluation image including a bar graph representing the error evaluation result and a correction reference line representing the correction reference for the demand forecasting model for each of the multiple customer groups. accumulating, in a database, teacher data including actual data on fluctuation factors of power demand and actual data on power demand for each of a plurality of customer groups of the power selling company;
generating a demand forecasting model representing a relationship between fluctuation factor data and electricity demand data for each of the plurality of customer groups based on a plurality of teacher data stored in the database;
acquiring prediction data of the fluctuation factors for each of the plurality of customer groups;
Calculating, for each of the plurality of customer groups, forecast data of electricity demand based on the forecast data of the fluctuation factors and the demand forecast model;
calculating forecast data of total electricity demand for the electricity seller based on the forecast data of electricity demand calculated for each of the plurality of customer groups;
Generating the demand forecast model includes:
generating an electricity demand forecasting model representing a relationship between fluctuation factor data including one fluctuation factor and electricity demand data for one customer group;
generating an electric power demand forecasting model representing a relationship between electric power demand data and fluctuation factor data including other fluctuation factors not included in the fluctuation factor data for the one customer group for another customer group;
The power demand forecasting method includes: accumulating, in a database, teacher data including actual data on fluctuation factors of power demand and actual data on power demand for each of a plurality of customer groups of the power selling company;
generating a demand forecasting model representing a relationship between fluctuation factor data and electricity demand data for each of the plurality of customer groups based on a plurality of teacher data stored in the database;
acquiring prediction data of the fluctuation factors for each of the plurality of customer groups;
Calculating, for each of the plurality of customer groups, forecast data of electricity demand based on the forecast data of the fluctuation factors and the demand forecast model;
calculating forecast data of total electricity demand for the electricity seller based on the forecast data of electricity demand calculated for each of the plurality of customer groups;
The plurality of customer groups are grouped according to differences in the variable factors affecting the electricity demand;
A method for forecasting electricity demand, wherein customers who are likely to have a similar relationship between the fluctuation factor data and the electricity demand data belong to the same customer group.

Images