Prediction device and prediction method

Abstract

To provide a prediction device and a prediction method capable of improving the accuracy of area price forecasts, and also to provide a prediction device and a prediction method capable of suppressing power procurement costs.SOLUTION: A prediction device includes a training unit that generates a trained model using, as training data, past area information in a target area and in an adjacent area, past interconnection-line information on an interconnection line between the target area and the adjacent area, and past area prices in the target area. Furthermore, the prediction device includes an inference unit that predicts a current area price in the target area by inputting current area information in the target area and in the adjacent area, and current interconnection-line information on the interconnection line, into the trained model.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present invention relates to a prediction device and a prediction method. [Background technology]

[0002] In recent years, due to the "full liberalization of the electricity market," there has been an increase in the number of new retail electricity suppliers known as "new power companies" (or "new power companies") entering the electricity market. New power companies may be able to supply electricity using renewable energy sources such as solar power generation and biomass power generation, but as this is dependent on factors such as the weather, they may not be able to provide a stable supply of electricity within their service areas.

[0003] Meanwhile, there is the Japan Electric Power Exchange (JEPX), a general incorporated association, which is an electricity exchange where domestic electricity is bought and sold. The electricity exchange has a market called the day-ahead market (or spot market) where electricity is delivered the next day. In the spot market, the trading price of electricity (price per kWh) is determined as the spot price in 30-minute increments. New power companies can make electricity demand plans for the next day, use the electricity exchange to procure electricity, and supply the procured electricity to consumers. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] "Indirect Transmission Rights," Agency for Natural Resources and Energy, October 30, 2017, Internet <https: / / www.meti.go.jp / shingikai / enecho / denryoku_gas / denryoku_gas / seido_kento / pdf / 013_03_00.pdf> Summary of the Invention [Problem to be solved by the invention]

[0005] The present disclosure provides a prediction device and a prediction method that can improve the accuracy of area price predictions and can reduce the cost of procuring electricity. [Means for solving the problem]

[0006] A prediction device according to a first aspect includes a learning unit that generates a trained model using past area information for a target area and an adjacent area adjacent to the target area, past interconnection line information for interconnection lines between the target area and the adjacent area, and past area prices for the target area as training data. The prediction device also includes an inference unit that inputs current area information for the target area and the adjacent area and current interconnection line information for the interconnection lines into the trained model to predict the current area price for the target area.

[0007] A prediction method according to a second aspect includes a step of generating a trained model using past area information for a target area and an adjacent area adjacent to the target area, past interconnection line information for interconnection lines between the target area and the adjacent area, and past area prices for the target area as training data. The prediction method also includes a step of inputting current area information for the target area and the adjacent area and current interconnection line information for the interconnection lines into the trained model to predict the current area price for the target area. [Effects of the Invention]

[0008] According to the present disclosure, it is possible to provide a prediction device and a prediction method that can improve the prediction accuracy of area prices and that can reduce the procurement costs of electricity. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram illustrating an example of buying and selling of electricity according to the first embodiment. [Figure 2]FIG. 2 is a diagram illustrating an example of a domestic power transmission and distribution network according to the first embodiment. [Figure 3] 3A and 3B are diagrams illustrating an example of a mechanism for determining a spot price according to the first embodiment. [Figure 4] FIG. 4 is a diagram illustrating an example of the configuration of a local power management system according to the first embodiment. [Figure 5] FIG. 5 is a diagram illustrating an example of the configuration of the local power management server according to the first embodiment. [Figure 6] FIG. 6 is a diagram illustrating an example of the prediction process according to the first embodiment. [Figure 7] FIG. 7 is a diagram illustrating an example of the configuration of a prediction device 140 according to the first embodiment. [Figure 8] FIG. 8 is a diagram illustrating an example of price congruence between areas according to the first embodiment. [Figure 9] FIG. 9 is a diagram illustrating an example of the relationship between the predicted value and the actual value of the spot price (for example, area price) according to the first embodiment. [Figure 10] FIG. 10 is a diagram illustrating an application example of the loss function according to the first embodiment. [Figure 11] FIG. 11 is a diagram illustrating an example of operation according to the first embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] [First embodiment] The prediction device according to the first embodiment will be described with reference to the drawings. In the description of the drawings, the same or similar parts are denoted by the same or similar reference numerals.

[0011] (Power purchase and sale) First, the buying and selling of electricity according to the first embodiment will be described.

[0012] FIG. 1 is a diagram showing an example of buying and selling electricity.

[0013] As shown in Figure 1, the electricity business can be divided into three categories: "power generation companies," "general electricity transmission and distribution companies," and "retail electricity companies."

[0014] A power generation company is a company that owns power generation facilities. Specifically, a power generation company is a company that generates electricity using power generation facilities such as hydroelectric power generation facilities, wind power generation facilities, and thermal power generation facilities. Power generation companies include not only major electric power companies (or regional electric power companies) such as Tokyo Electric Power Company and Kansai Electric Power Company, but also power generation companies that have newly entered the power generation business.

[0015] A general electricity transmission and distribution business operator is an entity that uses transmission lines and other transmission and distribution networks to deliver electricity generated by power generation businesses to the point where it is used by consumers. There are 10 supply areas in Japan, and a general electricity transmission and distribution business operator exists for each supply area.

[0016] A retail electricity supplier is a business that supplies electricity according to general demand. In addition to the 10 regional power companies in Japan, retail electricity suppliers also include new power suppliers (or new power companies) due to the "full liberalization of the electricity market." Retail electricity suppliers do not own power generation facilities or transmission and distribution networks. Therefore, retail electricity suppliers outsource the power generation business to power generation companies, and outsource the business of delivering electricity within the supply area to general transmission and distribution companies. Retail electricity suppliers pay a "power generation fee" to power generation companies, outsource the generation, and pay a "wheeling fee" to general transmission and distribution companies to use the transmission and distribution network. This enables retail electricity suppliers to supply electricity to contracted consumers (or customers).

[0017] One type of new power source that is attracting attention is the regional new power source (or regional new power company). A regional new power source is a company that operates a business supplying electricity within a specific region. Regional new power sources also belong to the category of retail electricity suppliers. As mentioned above, regional new power sources may supply electricity within a region using renewable energy sources such as solar power generation. However, since renewable energy sources are affected by factors such as the weather, a stable power supply may not be possible.

[0018] Therefore, regional new power companies also procure electricity using the spot market of the power exchange (Japan Electric Power Exchange). Regional new power companies can supply electricity generated by power generation companies to consumers within their region using the transmission and distribution network owned by general power transmission and distribution companies. This allows them to provide a stable supply of electricity to consumers within their region.

[0019] Generally, electricity cannot be stored in large quantities. Although technologies for storing electricity, such as batteries, have been developed in recent years, no storage technology capable of backing up the entire power supply has been developed. Therefore, production (or power generation) and consumption must be "same time and in the same amount." To achieve "same time and in the same amount," it is necessary to manage the supply and demand of electricity. If the supply and demand balance is disrupted, "same time and in the same amount" cannot be achieved, and large-scale blackouts may occur.

[0020] For this reason, power generation companies and electricity retailers need to achieve "same time and volume of planned demand" to ensure a stable supply of electricity. "Same time and volume of planned demand" means matching the amount of electricity demand with the amount of electricity supplied. To achieve this, power generation companies and electricity retailers create demand plans and submit them to the Organization for Cross-regional Coordination of Transmission Operators (OCCTO). OCCTO will manage the supply and demand of electricity throughout the country based on the demand plans. Regional new power companies also submit their demand plans to OCCTO and procure electricity for the amount of demand included in the demand plans through the power exchange.

[0021] As mentioned above, the power exchange has an electricity trading market called the spot market. In the spot market, electricity is traded in 30-minute increments at a trading price called the spot price. In the spot market, electricity trading for the next day closes at 10:00 a.m. on the day. Therefore, regional new power companies forecast the next day's demand volume within their region in 30-minute increments and conduct electricity trading based on the forecasted demand volume. Regional new power companies are also required to submit a demand plan detailing the demand volume to OCCTO. Furthermore, regional new power companies will calculate their own electricity procurement costs based on the forecasted spot price and the submitted demand plan.

[0022] The electricity is then transmitted within the area managed by the regional new power company. At this time, there may be cases where the amount of demand submitted by the regional new power company to OCCTO does not match the amount of supply actually supplied within the area. The difference between the amount of demand and the amount of actual supply is called an "imbalance." When an imbalance occurs, the general electricity transmission and distribution company takes responsibility and supplies electricity within the area so that the amount of demand matches the amount of supply. As a result, the regional new power company must pay an imbalance penalty (or imbalance fee) to the general electricity transmission and distribution company.

[0023] (spot price) In the spot market, participants (retail electricity suppliers) from all over the country submit bids to buy and sell electricity. The spot price is then determined by matching the lowest selling bid with the highest buying bid. The spot price represents the electricity rate per unit time in the spot market (yen / kWh). As mentioned above, the spot price is determined in 30-minute increments at the power exchange.

[0024] FIG. 2 is a diagram showing an example of a domestic power transmission and distribution network according to the first embodiment. As shown in FIG. 2, there are 10 regional power companies (power generation companies) in the country. In FIG. 2, the numbers in the circles represent the maximum power demand in each area as of August 2013. Also in FIG. 2, the numbers of the interconnection lines connecting each area represent the transmission capacity of the interconnection lines.

[0025] In Japan, due to the historical background of the power transmission and distribution network being developed in each of the 10 electricity company areas, the capacity of the interconnection lines is small compared to the maximum power demand in each area. Furthermore, 50% of the interconnection line capacity must be reserved for emergency response, and additional capacity must be reserved in the event that a connection contract to the power grid is concluded. The capacity remaining after subtracting the capacity reserved for emergency response and the capacity reserved in the event of a connection contract being concluded from the transmission capacity of the interconnection lines is sometimes called "available capacity."

[0026] 3A and 3B are diagrams illustrating an example of a mechanism for determining a spot price according to the first embodiment.

[0027] Figure 3A shows a case where demand in Area A exceeds supply by 200, and supply in Area B exceeds demand by 200, and there is also available capacity of 300 in the interconnection line. In this case, because there is available capacity of 300, it is possible to transmit the 200% of electricity that is oversupplied in Area B to Area A, where there is a supply shortage. This brings the supply and demand balance between Area A and Area B into agreement. With the supply and demand balance in agreement, it becomes possible to determine the spot price. If such transmission is possible in all areas of the country, the spot price can be unified nationwide. This nationwide unified spot price is called the "system price."

[0028] On the other hand, Figure 3B shows a case where Area A and Area B are in the same situation as Figure 3A, but the available capacity of the interconnection line is only "100". In this case, even if the available capacity of the interconnection line is only "100", and "100" of electricity is transmitted from Area B to Area A, the supply shortage in Area A will not be resolved, and the oversupply in Area B will not be resolved either. In this case, the spot price in Area A will be determined in a state of shortage, and the spot price in Area B will be determined in a state of oversupply. In this case, the spot price in Area A, which is experiencing a supply shortage, may be higher than the spot price in Area B, which is experiencing an oversupply. When spot prices differ between areas, these spot prices are called "area prices". Furthermore, a state in which spot prices differ between areas is called a state in which "market fragmentation" occurs.

[0029] As such, spot prices consist of system prices (Figure 3A) and area prices (Figure 3B). In particular, there are cases where the difference between areas in the amount of electricity generated by renewable energy sources such as wind power becomes larger than a certain level. As a result, in some areas, there may be an oversupply of electricity due to renewable energy sources, causing the area price to approach "0."

[0030] (Prediction according to the first embodiment) Generally, electricity demand is highest in the afternoon in summer and in the morning in winter. For this reason, it is assumed that spot prices are also predictable depending on the season and time of day.

[0031] However, there are cases where spot prices rise due to some reason. For regional new power companies, electricity costs are also a management risk. Therefore, for example, if regional new power companies can predict a rise in spot prices, they will be able to adjust their electricity procurement at that time, suppress electricity costs, and avoid management risks.

[0032] Therefore, in the first embodiment, an example will be described in which an AI model is used to predict area prices within an area.

[0033] Specifically, first, a prediction device generates a trained model using, as training data, past area information for a target area and an adjacent area adjacent to the target area, past interconnection line information for interconnection lines between the target area and the adjacent area, and past area prices in the target area. In the first embodiment, the trained model is generated using not only the area information for adjacent areas and the area prices of the target area, but also the interconnection line information for the interconnection lines between the target area and the adjacent area.

[0034] Second, current area information for adjacent areas and current interconnection line information for interconnection lines are input into the trained model to predict the current area price for the target area.

[0035] In this way, in the first embodiment, the area price of the target area can be predicted by taking into account not only the area information of adjacent areas but also the interconnection line information. Therefore, compared to when the interconnection line information is not taken into account, it is possible to predict the area price with higher accuracy. Therefore, in the first embodiment, it is possible to improve the accuracy of the area price prediction. Furthermore, by improving the accuracy of the area price prediction, it is also possible to predict when the area price will rise, and in such cases, it is possible to reduce the amount of electricity procured, thereby making it possible to suppress the power procurement cost.

[0036] The area may be a service area to which a general electricity transmission and distribution company supplies electricity. Alternatively, the area may be an area managed by a regional power company. Alternatively, the area may be an area including a predetermined range on a map.

[0037] (Regional power management system according to the first embodiment) Next, the local power management system according to the first embodiment will be described.

[0038] FIG. 4 is a diagram illustrating an example of the configuration of the local power management system 10. As shown in FIG.

[0039] 4, the local power management system 10 includes a local power management server 100, a facility 200, and a network 300. The local power management system 10 may also be called an Area Energy Management System (AEMS).

[0040] The local power management server 100 is a server managed by a local power supplier (electricity retailer). The local power management server 100 manages the power of facilities 200 installed within a region via a network 300. The local power management server 100 may transmit a control message to an EMS (Energy Management System) 230 installed in the facility 200, instructing control of a distributed power source 210 installed in the facility 200. The control message may be, for example, a power flow control message requesting control of a power flow (the flow of power from the power grid 250 to the facility 200), or a reverse power flow control message requesting control of a reverse power flow (the flow of power from the facility 200 to the power grid 250). Alternatively, the control message may be a power control message that controls the operating state of the distributed power source 210.

[0041] The regional power management server 100 according to the first embodiment acquires area information, grid line information, and area prices from an external server 400 via a network 300. The regional power management server 100 uses the area information, grid line information, and area prices to generate a trained model using a machine learning function. The regional power management server 100 then uses the generated trained model to predict area prices from the area information and grid line information. The regional power management server 100 according to the first embodiment also creates a demand plan and transmits data related to the demand plan to an OCCTO management server (external server 400) managed by OCCTO, and uses the predicted area prices to procure power from a power exchange management server (external server 400) managed by a power exchange.

[0042] The facility 200 is a facility installed within the area managed by the local power management system 10. In Fig. 4, an example is shown in which a facility 200A and a facility 200B are installed as the facility 200, but the number of facilities 200 installed within the area may be one, or three or more. The facility 200 includes a distributed power source 210, a load device 220, and an EMS 230.

[0043] The distributed power source 210 is a device that generates power. The distributed power source 210 may be a device that generates power using natural energy such as sunlight, wind power, hydroelectric power, or geothermal power, or may be a device that generates power using biomass (organic matter) such as animals or plants as raw materials. Biomass energy may also be included in natural energy. The distributed power source 210 may be a solar power generation device (or solar cell device), wind power generation device, hydroelectric power generation device, geothermal power generation device, biomass power generation device, or the like.

[0044] The load device 220 is a device that consumes power, such as an air conditioner, a lighting device, or an AV (Audio and Video) device.

[0045] The EMS 230 is a device that manages the power of the facility 200. The EMS 230 may control the operating states of the distributed power sources 210 and the load devices 220.

[0046] Here, the EMS 230 receives power generation amount data indicating the amount of power generated by the distributed power source 210 from the distributed power source 210. The power generation amount data may indicate the output power of the distributed power source 210. The EMS 230 may periodically receive the power generation amount data from the distributed power source 210. The EMS 230 transmits the power generation amount data to the local power management server 100. The EMS 230 may periodically transmit the power generation amount data to the local power management server 100.

[0047] Communication between the local power management server 100 and the EMS 230 may be performed in accordance with a first protocol. The first communication protocol may be, for example, a protocol conforming to Open ADR (Automated Demand Response) or a proprietary dedicated protocol. In the first embodiment, communication between the EMS 230 and the distributed power sources 210 may be performed in accordance with a second protocol different from the first protocol. The second communication protocol may be, for example, a protocol conforming to ECHONET Lite (registered trademark), SEP (Smart Energy Profile) 2.0, KNX, or a proprietary dedicated protocol.

[0048] The network 300 connects the local power management server 100 and the facility 200. The network 300 may be the Internet (registered trademark) or a dedicated line such as a VPN (Virtual Private Network).

[0049] An external server 400 is connected to the local power management system 10 via a network 300.

[0050] First, as described above, the external server 400 may be an OCCTO management server managed by OCCTO. In this case, the local power management server 100 acquires interconnection line information and wide-area reserve margin information from the OCCTO management server. The wide-area reserve margin information is, for example, information that indicates an index of supply relative to power demand. The wide-area reserve margin information is an example of area information.

[0051] Second, as described above, the external server 400 may be a power exchange management server managed by the power exchange. In this case, the regional power management server 100 acquires price information such as system prices and area prices from the power exchange management server. The price information may be included in the area information, but in the following description, it will be distinguished from the area information.

[0052] Third, the external server 400 may be a server managed by the Japan Meteorological Agency. In this case, the local power management server 100 acquires weather forecast information representing information related to weather from the Japan Meteorological Agency management server. The weather forecast information is an example of area information.

[0053] Fourth, the external server 400 may be a general electricity transmission and distribution company management server managed by a general electricity transmission and distribution company. In this case, the regional power management server 100 acquires power generation performance information indicating power generation performance from the general electricity transmission and distribution company management server. The power generation performance information is also an example of area information.

[0054] (Example of a regional power management server configuration) Next, a configuration example of the local power management server 100 according to the first embodiment will be described.

[0055] FIG. 5 is a diagram illustrating an example of the configuration of the local power management server 100. As shown in FIG.

[0056] As shown in FIG. 5, the local power management server 100 includes an interface unit 110, a control unit 120, and a storage unit 150.

[0057] The interface unit 110 communicates with the EMS 230 of the facility 200 via the network 300 under the control of the control unit 120. In this case, the interface unit 110 communicates with the EMS 230 using data (or packets) of a first communication protocol. Furthermore, the interface unit 110 communicates with the external server 400 via the network 300 under the control of the control unit 120. In this case, the interface unit 110 may communicate with the external server 400 using data (or packets) that comply with a dedicated protocol established between the interface unit 110 and the external server 400. The dedicated protocol may be the Internet Protocol (IP), and in this case, the interface unit 110 communicates with the external server 400 using IP packets.

[0058] The control unit 120 controls the regional power management server 100. The control unit 120 includes a supply and demand management unit 130 and a customer management unit 135. The control unit 120 may include at least one memory and at least one processor electrically connected to the memory. In this case, the memory stores programs executed by the processor and information used in processing by the processor. The processor may implement the functions of the supply and demand management unit 130 and the customer management unit 135 by reading and executing the programs stored in the memory.

[0059] The supply and demand management unit 130 includes a power data collection unit 131, a prediction unit 140, a power procurement unit 132, a power plan submission unit 133, and a supply and demand monitoring and adjustment unit 134. As described above, the control unit 120 (a processor included in the control unit 120) may execute a program to realize the functions of the power data collection unit 131, the prediction unit 140, the power procurement unit 132, the power plan submission unit 133, and the supply and demand monitoring and adjustment unit 134.

[0060] The power data collection unit 131 collects power data from the general power transmission and distribution company management server (external server 400) via the interface unit 110. The power data may include, for example, the amount of power supplied by the general power transmission and distribution company to the area every 30 minutes.

[0061] The prediction unit 140 predicts the area price of the target area using the area information and the interconnection line information. The prediction process performed by the prediction unit 140 will be described in detail later.

[0062] The power procurement unit 132 procures power by accessing the power exchange management server via the interface unit 110. For example, the power procurement unit 132 procures power in the spot market by transmitting and receiving information related to power procurement to and from the power exchange management server. The power procurement unit 132 may procure power using the area price predicted by the prediction unit 140.

[0063] The power plan submission unit 133 accesses the OCCTO management server via the interface unit 110 and transmits information about the power plan. The power plan submission unit 133 may generate information about power demand using the area price predicted by the prediction unit 140. The information about power demand includes, for example, data about the demand plan.

[0064] The supply and demand monitoring and adjusting unit 134 adjusts the supply and demand situation of electricity in the facility 200 via the interface unit 110. The supply and demand monitoring and adjusting unit 134 may adjust the supply and demand situation based on power generation amount data received from the EMS 230, or may adjust the supply and demand situation by transmitting a control message to the EMS 230 instructing control of the dispersed power source 210.

[0065] The customer management unit 135 manages customers in the area who have contracted with the local new power company. The customer management unit 135 performs, for example, customer registration processing, customer information management processing, electricity usage fee calculation processing, power fee billing processing, and fee payment management processing.

[0066] The storage unit 150 stores various types of information under the control of the control unit 120.

[0067] 5, the prediction unit 140 may be executed as a function in the control unit 120, or the function of the prediction unit 140 may be executed as a device (for example, a prediction device) separate from the local power management server 100. When the prediction unit 140 is executed as a prediction device, the prediction device 140 may be connected to the local power management server 100 via a network 300.

[0068] (Area price prediction processing) Here, the prediction process using machine learning performed by the prediction unit 140 will be described.

[0069] FIG. 6 is a diagram illustrating an example of the prediction process according to the first embodiment.

[0070] Figure 6 shows an example of predicting area prices for the "Tokyo" area. In Figure 6, each area represents an area managed by a local electric power company.

[0071] The prediction process according to the first embodiment uses graph convolutional neural networks (GCNs) (hereinafter referred to as "GCNs"). GCNs are a deep learning method that uses graphs. GCNs are also neural networks that are generated by quantifying the features of each node through convolution without changing the structure of the graph itself.

[0072] GCN uses an adjacency matrix and a feature matrix as input.

[0073] An adjacency matrix is ​​a matrix that represents the connection state between nodes. In the adjacency matrix shown in Figure 6, "Tohoku" and "Tokyo" are adjacent, so the value is "1," and "Tohoku" and "Chubu" are not adjacent, so the value is "0." The values ​​in the adjacency matrix are expressed as "1" for adjacent areas and "0" for non-adjacent areas.

[0074] The feature matrix is ​​a matrix that represents the feature quantities (or feature vectors) of each node. The feature matrix includes area information and area prices for each area as features. For example, feature quantity #1 represents "temperature" among the area information, feature quantity #2 represents the area price one day earlier, and feature quantity #2 represents the area demand among the area information. In this case, in the feature matrix, the temperature (feature quantity #1) for "Tohoku" is "7.0" (°C), the area price (feature quantity #2) for "Tohoku" one day earlier is "6.0" (yen / kWh), and the area demand (feature quantity #3) for "Tohoku" is "6.0" (yen / kWh). Note that for the sake of convenience, the feature matrix shown in FIG. 6 shows an example in which the number of features is "3," but in the first embodiment, a greater number of features are used.

[0075] In GCN, if the adjacency matrix is ​​A, the feature matrix is ​​X, and the weighting matrix is ​​W, the determinant of AXW is calculated. Here, we focus on the calculation results of AX. For the first row of the calculation results of AX ("Tohoku"), the feature values ​​of the non-adjacent "Chubu" are not included, but the result includes the feature values ​​of the adjacent "Tokyo". For the second row of the calculation results of AX ("Tokyo"), the result includes the two feature values ​​of the adjacent "Tohoku" and "Chubu".

[0076] In this way, GCN can obtain results that take into account the features of adjacent nodes by calculating AX. The results obtained by GCN are called "latent matrices," and the values ​​contained in the "latent matrices" are called "latent variables." If the latent matrix is ​​H, GCN can be expressed as follows: H=AXW (1) The calculation of equation (1) may be performed multiple times.

[0077] As shown in Figure 6, in the adjacency matrix, the area itself, such as "Tohoku" and "Tohoku," is represented as "0." Therefore, the latent matrix H will not include the feature of the area itself. Therefore, in GCN, Equation (1) may be used, where A = A + I (I is the unit matrix).

[0078] In the first embodiment, a feature matrix including interconnection line information is added to the GCN. As described above, the available capacity of the interconnection line may be taken into consideration in the process of determining the area price. Therefore, by executing the GCN while taking into account interconnection line information related to the interconnection line, such as available capacity, it becomes possible to predict area prices that are in line with the actual determination process. Therefore, it is expected that the accuracy of area price prediction will be improved compared to when the interconnection line information is not taken into consideration.

[0079] Here, specific examples of feature quantities included in the feature matrix will be described. Specific examples of feature quantities are summarized in the table below.

[0080] [Table 1] In the table, "area" in "location of use" represents the feature amount in the area, and "interconnection line" represents the feature amount in the interconnection line.

[0081] Details of each piece of information are as follows:

[0082] "Lower, middle, upper and total cloud cover": Represents the cloud cover at each layer and the total cloud cover at the predetermined lower, middle and upper layers.

[0083] "System Price" and "Area Price": For example, these represent the prices for every 30 minutes.

[0084] "Power generation performance": For example, this indicates the amount of power generated using power generation facilities in each area.

[0085] "Photovoltaic power generation performance": For example, it indicates the amount of power generated using photovoltaic power generation within the area. The power generation performance information may indicate, for example, the amount of power generated for each hour.

[0086] Details of each piece of information included in the wide-area reserve margin information are as follows:

[0087] "Wide-area block demand": This refers to the amount of demand in a wide-area block. A wide-area block is an area without interconnection line congestion, and includes one or more areas.

[0088] "Wide-area block supply capacity": Indicates the amount of electricity that can be supplied in a wide-area block.

[0089] "Wide-area block reserve capacity": Represents the amount obtained by subtracting "wide-area block supply capacity" from "wide-area block demand."

[0090] "Wide-area reserve margin": Indicates the ratio of "wide-area block reserve capacity" to "wide-area block demand."

[0091] "Area demand": Represents the amount of demand in an area.

[0092] "Area supply capacity": Indicates the amount of electricity that can be supplied in the area.

[0093] "Area reserve capacity": Represents the amount obtained by subtracting "area supply capacity" from "area demand."

[0094] Details of each piece of information included in the interconnection line information are as follows:

[0095] "Direction": A specific direction (for example, from north to south or from east to west) is "1", and the opposite direction is "0".

[0096] "Planned power flow": This refers to the capacity managed by OCCTO as the sum of the capacities registered by interconnection line users.

[0097] "Wide-area adjustment frame": Represents the amount of electricity required to match the supply and demand of electricity between areas.

[0098] "Margin": This represents the capacity managed by OCCTO as part of the interconnection line's operational capacity to receive electricity from other supply areas via the interconnection line in emergency situations such as when the power grid is abnormal.

[0099] "Operating capacity": This represents the capacity of the interconnection lines operated by OCCTO.

[0100] "Number of transmission NG information cases": Indicates the number of pieces of information indicating that power transmission using interconnection lines is NG.

[0101] "Total capacity of transmission NG information": Represents the total capacity when power transmission using interconnection lines is NG.

[0102] (Example of prediction device configuration) Next, a description will be given of an example configuration of the prediction device 140 according to the first embodiment. As described above, the prediction device 140 may be a prediction unit.

[0103] FIG. 7 is a diagram illustrating an example of the configuration of the prediction device 140. As shown in FIG.

[0104] As shown in FIG. 7, the prediction device 140 includes a data acquisition unit 141, a learning unit 142, an evaluation unit 143, a trained model storage unit 144, an inference unit 145, and a data storage unit 146.

[0105] The data acquisition unit 141 acquires past area information, past area prices, and past interconnection line information from the external server 400 via the interface unit 110. The data acquisition unit 141 outputs the past area information, past area prices, and past interconnection line information to the learning unit 142.

[0106] The learning unit 142 generates a GCN-based neural network model, i.e., a trained model, using past area information, past area prices, and past interconnection line information. The area information includes area information for the target area and area information for adjacent areas adjacent to the target area. The learning unit 142 creates a trained model using area information, area prices, and interconnection line information at a certain point in time in the past, and then updates the trained model using area information, area prices, and interconnection line information at the next point in time. The learning unit 142 generates (or updates) the trained model using area information, area prices, and interconnection line information over a predetermined period of time.

[0107] FIG. 8 is a diagram illustrating an example of price congruence between areas according to the first embodiment. As shown in FIG. 8, for example, if the area prices become the same between "Hokuriku" and "Kyushu" at a certain point in time, a situation will arise in which the area prices will also become the same in areas ("Kansai" and "Chugoku") through which interconnection lines pass between "Hokuriku" and "Kyushu." This situation in which area prices become the same between distant areas and the area prices through which interconnection lines pass between the areas are also the same is called "price congruence." The learning unit 142 uses area information and area price information to generate a trained model over a certain period of time, making it possible to generate a trained model that takes into account the "price congruence" situation.

[0108] Returning to FIG. 7, the learning unit 142 outputs the learned model to the evaluation unit 143.

[0109] The evaluation unit 143 evaluates the trained model generated by the learning unit 142. Specifically, the evaluation unit 143 may evaluate the trained model as follows. That is, the evaluation unit 143 inputs past area information and past interconnection line information into the trained model, and causes it to predict the area price as a predicted value. Then, the evaluation unit 143 uses the predicted area price (predicted value) and the past area price (actual measured value) to perform evaluation using a loss function.

[0110] Here, the evaluation unit 143 evaluates the trained model using a loss function that places more importance on error evaluation in areas where prediction is less accurate than in other areas.

[0111] FIG. 9 is a diagram illustrating an example of the relationship between the predicted value and the actual measured value of the spot price (e.g., area price) according to the first embodiment. As shown in FIG. 9, when the actual measured value of the spot price is equal to or greater than the first reference threshold at a certain point in time, the predicted value of the spot price at that point in time is significantly different from the actual measured value. Also, when the actual measured value of the spot price is equal to or less than the second reference threshold (second reference threshold<first reference threshold) at a certain point in time, the predicted value of the spot price at that point in time is significantly different from the actual measured value. In this way, there are cases where the predicted value of the spot price and the actual measured value differ significantly by more than a certain amount.

[0112] 10 is a diagram showing an application example of the loss function according to the first embodiment. In the first embodiment, the region where the actual spot price value and the predicted value differ significantly by more than a certain amount when the actual spot price is in the top 20% and the bottom 20% is defined as a region where prediction is difficult to make. The evaluation unit 143 then evaluates the trained model using the loss function shown below, which places emphasis on this region.

[0113]

number

[0114]

number

[0115] Returning to FIG. 7, the evaluation unit 143 stores the trained model after evaluation in the trained model storage unit 144.

[0116] The inference unit 145 predicts area prices using a trained model read from the trained model storage unit 144. Specifically, the inference unit 145 inputs current area information for the target area and adjacent areas and current interconnection line information for the interconnection lines into the trained model, and predicts the current (or future) area prices for the target area by inference. As described above, the trained model is trained using not only the area information for the target area and adjacent areas but also the interconnection line information for the interconnection lines between the target area and adjacent areas. This makes it possible to predict area prices with higher accuracy than when interconnection line information is not used. The inference unit 145 stores the predicted (or inferred) area prices in the data storage unit 146. The area prices stored in the data storage unit 146 are read out by the power procurement unit 132, for example, and used for power procurement.

[0117] (Operation example according to the first embodiment) Next, an example of operation according to the first embodiment will be described.

[0118] Fig. 11 is a diagram illustrating an example of an operation according to the first embodiment. Fig. 11 illustrates an example of an operation performed by the prediction device 140 (or the prediction unit 140).

[0119] 11, when the process starts in step S10, the data acquisition unit 141 acquires data to be used for learning in step S11. The data acquisition unit 141 acquires area information, interconnection line information, and area prices from the external server 400 as data.

[0120] In step S12, the learning unit 142 performs learning by the GCN using the data acquired by the data acquisition unit 141. The learning unit 142 generates a trained model.

[0121] In step S13, the evaluation unit 143 uses an evaluation function to evaluate the trained model generated by the learning unit 142. The evaluation unit 143 stores the trained model after evaluation in the trained model storage unit 144.

[0122] In step S14, the inference unit 145 uses the trained model read from the trained model storage unit 144 to predict the area price of the target area.

[0123] In step S15, the inference unit 145 stores the predicted area price in the data storage unit 146.

[0124] Then, in step S16, the prediction device 140 ends the series of processes.

[0125] (Another example 1 according to the first embodiment) In the first embodiment, weather forecast information, power generation performance information, and wide-area reserve margin information are used as examples of area information. However, at least a part of the weather forecast information, power generation performance information, and wide-area reserve margin information may be used as area information.

[0126] In the first embodiment, ten types of weather forecast information are used as an example, but at least one type of weather forecast information may be used. Furthermore, for the power generation performance information and the wide-area reserve margin information, at least one of two types of power generation performance information may be used, or at least one of eight types of wide-area reserve margin information may be used.

[0127] (Another example 2 according to the first embodiment) In the first embodiment, an example has been described in which a GCN is used in the prediction process for predicting area prices, but the prediction process is not limited to this. A machine learning method other than a GCN may be used in the prediction process as long as it can utilize machine learning using a graph (or a neural network using a graph (GNN: Graph Neural Network)). Examples of such machine learning methods include GAT (Graph Attention Networks), GraphSAGE, and NAS (Neural Architecture Search). GAT is a machine learning method that combines, for example, a GNN and an attention layer (a layer that can dynamically set which part of the input data to focus on). GraphSAGE is an inductive machine learning method that uses, for example, attribute information of nodes to represent data that has never been seen before. NAS is a method that combines, for example, machine learning methods using graphs.

[0128] [Other embodiments] A program may be provided that causes a computer to execute each process performed by the above-mentioned devices (regional power management server 100 and prediction device 140). The program may be recorded on a computer-readable medium. The computer-readable medium can be used to install the program on a computer. Here, the computer-readable medium on which the program is recorded may be a non-transitory recording medium. The non-transitory recording medium is not particularly limited, and may be, for example, a recording medium such as a CD-ROM or a DVD-ROM. Furthermore, circuits that execute each process performed by the above-mentioned devices (regional power management server 100 and prediction device 140) may be integrated, and the device may be configured as a semiconductor integrated circuit (chipset, SoC).

[0129] Although the embodiments have been described in detail above with reference to the drawings, the specific configuration is not limited to the above, and various design changes can be made without departing from the spirit of the invention. Furthermore, the operation examples and processes can be combined within a consistent range.

[0130] (Addendum) The above can be summarized as follows:

[0131] (Appendix 1) a learning unit that generates a trained model using, as training data, past area information in a target area and an adjacent area adjacent to the target area, past interconnection line information on an interconnection line between the target area and the adjacent area, and past area prices in the target area; an inference unit that inputs current area information in the target area and the adjacent area and current interconnection line information in the interconnection line into the trained model and predicts a current area price in the target area; A prediction device having the following.

[0132] (Appendix 2) The interconnection line information includes a feature amount related to the interconnection line between the target area and the adjacent area. 2. The prediction device of claim 1.

[0133] (Appendix 3) The area information includes at least one of weather forecast information indicating information related to weather, power generation performance information indicating power generation performance, and wide-area reserve margin information indicating an index of supply to demand for power in a wide-area block. 3. The prediction device according to claim 1 or 2.

[0134] (Appendix 4) The learning unit generates the trained model by using an adjacency matrix representing a connection relationship between the target area and the adjacent area, a first feature matrix including the past area information and the past area prices, and a second feature matrix including the interconnection line information, and employs machine learning using a graph. 4. The prediction device according to claim 1,

[0135] (Appendix 5) The method further includes an evaluation unit that evaluates the trained model using a weighted loss function when the area price is equal to or greater than a first reference threshold or when the area price is equal to or less than a second reference threshold (second reference threshold<first reference threshold). 5. A prediction device according to any one of Supplementary Note 1 to Supplementary Note 4.

[0136] (Appendix 6) The learning unit generates the trained model taking into consideration price congruence in which, when area prices become the same in a first area and a second area that are not adjacent to each other, area prices of areas through which a connecting line passes between the first area and the second area become the same. 6. A prediction device according to any one of Supplementary Note 1 to Supplementary Note 5.

[0137] (Appendix 7) generating a trained model using, as training data, past area information in a target area and an adjacent area adjacent to the target area, past interconnection line information on an interconnection line between the target area and the adjacent area, and past area prices in the target area; and inputting current area information in the target area and the adjacent area and current interconnection line information in the interconnection line into the trained model to predict the current area price in the target area. Forecasting methods. [Explanation of symbols]

[0138] 10: Regional power management system 100: Regional power management server 110: Interface unit 120: Control unit 130: Supply and demand management unit 131: Power data collection unit 132: Power Procurement Department 133: Power Plan Submission Department 134: Supply and demand monitoring and adjustment unit 140: Forecasting unit (forecasting device) 141: Data acquisition unit 142: Learning unit 143: Evaluation unit 144: Learned model storage unit 145: Inference unit 146: Data storage unit 150: Storage unit 400: External server

Claims

1. a learning unit that generates a trained model using, as training data, past area information in a target area and an adjacent area adjacent to the target area, past interconnection line information on an interconnection line between the target area and the adjacent area, and past area prices in the target area; an inference unit that inputs current area information in the target area and the adjacent area and current interconnection line information of the interconnection line into the trained model and predicts a current area price in the target area; A prediction device having the following.

2. The interconnection line information includes a feature amount related to the interconnection line between the target area and the adjacent area. The prediction device according to claim 1 .

3. The area information includes at least one of weather forecast information indicating information related to weather, power generation performance information indicating power generation performance, and wide-area reserve margin information indicating an index of supply to demand for power in a wide-area block. The prediction device according to claim 1 .

4. The learning unit generates the trained model by using an adjacency matrix representing a connection relationship between the target area and the adjacent area, a first feature matrix including the past area information and the past area prices, and a second feature matrix including the interconnection line information, and employs machine learning using a graph. The prediction device according to claim 1 .

5. The method further includes an evaluation unit that evaluates the trained model using a weighted loss function when the area price is equal to or greater than a first reference threshold or when the area price is equal to or less than a second reference threshold (second reference threshold<first reference threshold). The prediction device according to claim 1 .

6. The learning unit generates the trained model taking into consideration price congruence in which, when area prices become the same in a first area and a second area that are not adjacent to each other, area prices of areas through which a connecting line passes between the first area and the second area become the same. The prediction device according to claim 1 .

7. generating a trained model using, as training data, past area information in a target area and an adjacent area adjacent to the target area, past interconnection line information on an interconnection line between the target area and the adjacent area, and past area prices in the target area; and inputting current area information in the target area and the adjacent area and current interconnection line information in the interconnection line into the trained model to predict the current area price in the target area. Forecasting methods.

Images