Prediction device and prediction method

Abstract

This prediction device comprises: a verification prediction unit that predicts electricity prices in a predetermined verification period, on the basis of each of a plurality of trained prediction models; a model selection unit that selects a prediction model having high accuracy for predicting electricity prices among the prediction models; and a target prediction unit that predicts electricity prices in a prediction target period, on the basis of the selected prediction model.

Description

Prediction device and prediction method

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

[0002] There is a wholesale electricity trading market where electricity retailers and power generators trade electricity. In the wholesale electricity trading market, electricity generators bid the amount of electricity they will sell and the selling price, and retailers bid the amount of electricity they will buy and the purchase price, and an electricity transaction is concluded. In the wholesale electricity trading market, the agreed price at which an electricity transaction is concluded is also called the electricity market price.

[0003] Techniques for predicting electricity market prices are known. For example, Patent Literature 1 discloses an electricity market price prediction device that predicts the electricity market price on a future target prediction date based on an estimated supply amount of electricity generated by renewable energy on that target prediction date.

[0004] Japanese Patent Application Laid-Open No. 2019-96164

[0005] However, the accuracy of electricity price predictions in conventional technologies can be improved. For example, because electricity prices are determined by a complex combination of various factors, it is difficult to accurately predict electricity prices based solely on the supply and demand relationship.

[0006] In view of the above technical problems, one aspect of the present invention aims to accurately predict electricity prices.

[0007] A prediction device according to a first aspect of the present invention comprises a verification prediction unit that predicts electricity prices for a specified verification period based on each of a plurality of trained prediction models, a selection unit that selects a prediction model from the plurality of prediction models that has high prediction accuracy for electricity prices, and a target prediction unit that predicts electricity prices for a specified prediction period based on the selected prediction model.

[0008] A prediction device according to a second aspect of the present invention includes an acquisition unit that acquires a fuel price indicating the price of fuel used for power generation, and a prediction unit that predicts electricity prices by inputting explanatory variables including the fuel price into a trained prediction model.

[0009] FIG. 1 is a block diagram showing an example of the overall configuration of an electricity price prediction system. FIG. 2 is a block diagram showing an example of the hardware configuration of a computer. FIG. 3 is a block diagram showing an example of the functional configuration of a prediction device in a first embodiment. FIG. 4 is a flowchart showing an example of a prediction method in the first embodiment. FIG. 5 is a block diagram showing an example of the functional configuration of a prediction device in a second embodiment. FIG. 6 is a flowchart showing an example of a prediction method in the second embodiment. FIG. 7 is a scatter plot showing an example of prediction accuracy for each index price. FIG. 8 is a line graph showing an example of prediction accuracy for each index price.

[0010] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In this specification and drawings, components having substantially the same functional configurations are designated by the same reference numerals, and redundant description will be omitted.

[0011] [First embodiment] A first embodiment of the present invention is an electricity price prediction system for predicting electricity prices. The electricity price prediction system in the first embodiment has a function of predicting electricity prices based on a trained prediction model.

[0012] The electricity price prediction system is capable of predicting any electricity price as long as it is determined by a method based on legal grounds. The electricity price determined by a method based on legal grounds does not mean that the electricity price is determined directly by law, but rather that the electricity price is determined by the actions of traders in accordance with certain rules established by law.

[0013] The electricity price may be, for example, the electricity market price. The electricity market price is the contracted price at which an electricity transaction is concluded in the electricity trading market. The electricity trading market may be a wholesale electricity trading market or a baseload market. The wholesale electricity trading market may be a spot market, an hourly market, or a forward market. The electricity trading market or wholesale electricity trading market is not limited to these, and may be any market where electricity transactions take place. Furthermore, as another example, the electricity price may be an imbalance charge unit price. The imbalance charge unit price is a unit price at which a general electricity transmission and distribution company adjusts the surplus or shortage of electricity in actual supply and demand. In this embodiment, a configuration for predicting the electricity market price will be described as an example of the electricity price.

[0014] The electricity trading market is provided as a place for electricity trading between retail electricity suppliers and power generators. In the electricity trading market, electricity generators bid the amount of electricity they will sell and the selling price, and retail electricity suppliers bid the amount of electricity they will buy and the purchasing price, whereby electricity trading takes place.

[0015] Forecasting electricity market prices is important for optimizing market bidding and creating appropriate energy resource operation plans. Improving the accuracy of electricity market price forecasts is beneficial for both electricity retailers and power generators.

[0016] However, it is difficult to predict the electricity market price. For example, in the wholesale electricity trading market, the electricity market price is determined at the balance between the selling bids from power generation companies and the buying bids from electricity retailers. Therefore, the electricity market price fluctuates greatly depending on the volume or price of the bids for buying and selling.

[0017] Furthermore, wholesale electricity markets sometimes use a contracting method known as the blind single price auction method. In this method, bidding trends on the day are not made public, so retail electricity suppliers and power generation companies are unable to grasp the actual state of the supply and demand curve on the day of the auction.

[0018] The present embodiment aims to accurately predict electricity market prices. In this embodiment, in order to improve the prediction accuracy of electricity market prices, at least one of teacher data, a prediction algorithm, and an explanatory variable can be selected. In one aspect, this embodiment can provide a prediction device that can accurately predict electricity market prices.

[0019] <Overall Configuration> The overall configuration of the electricity price prediction system in the first embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing an example of the overall configuration of the electricity price prediction system.

[0020] 1, the electricity price prediction system 1000 includes a prediction device 10 and a terminal device 20. The prediction device 10 and the terminal device 20 are connected to each other so as to be able to communicate data with each other via a communication network N1 such as a local area network (LAN) or the Internet.

[0021] The electricity price prediction system 1000 is connected to an external data source 30 via a communication network N1 so as to be able to communicate data therewith. The external data source 30 includes at least an electricity market system 31 and an LNG (Liquefied Natural Gas) market system 32. The external data source 30 may also include a weather information system 33 or an exchange market system 34.

[0022] The external data source 30 is not limited to these and may include other data sources. For example, the external data source 30 may include other data sources that accumulate data acquired from the electricity market system 31, the LNG market system 32, the weather information system 33, or the foreign exchange market system 34, and process and provide the data.

[0023] The prediction device 10 is an information processing device such as a personal computer, workstation, or server that predicts the electricity market price formed in the wholesale electricity trading market. The prediction device 10 acquires performance data from an external data source 30 and generates multiple prediction models based on the performance data. The prediction device 10 selects a prediction model with high prediction accuracy from the multiple prediction models and predicts the electricity market price using the selected prediction model. The prediction device 10 transmits the predicted result of the electricity market price to the terminal device 20.

[0024] The terminal device 20 is an information processing terminal such as a personal computer, a smartphone, or a tablet terminal operated by a user of the electricity price prediction system 1000. The terminal device 20 receives the prediction result of the electricity market price from the prediction device 10 and presents the prediction result to the user.

[0025] The electricity market system 31 is an information processing system that provides a wholesale electricity trading market. The wholesale electricity trading market is a market where electricity trading takes place between retail electricity suppliers and power generation suppliers. An example of a wholesale electricity trading market is the Japan Electric Power Exchange (JEPX). In the wholesale electricity trading market, multiple electricity market prices may be formed, for example, by region or by power generation method. The electricity market system 31 accumulates actual values ​​of the electricity market prices formed in the wholesale electricity trading market.

[0026] The LNG market system 32 is an information processing system that provides an LNG market. The LNG market is a market where LNG futures trading takes place. An example of an LNG market is the Tokyo Commodity Exchange. The LNG market system 32 accumulates actual values ​​of index prices that are statistically processed from futures prices formed in the LNG market. An example of an index price is Platts JKM (Japan Korea Marker).

[0027] The weather information system 33 is an information processing system that provides weather forecast data. The weather information system 33 stores weather forecast data indicating weather forecasts that have been announced in the past. The weather information system 33 may also store weather record data indicating actually observed weather information.

[0028] The foreign exchange market system 34 is an information processing system that provides a foreign exchange market. The foreign exchange market is a market where different currencies are exchanged. An example of a foreign exchange market is the Tokyo Foreign Exchange Market. The foreign exchange market system 34 accumulates actual values ​​of exchange rates formed in the foreign exchange market. An example of an exchange rate is the US dollar / Japanese yen rate.

[0029] The overall configuration of the electricity price prediction system 1000 shown in FIG. 1 is one example, and various system configuration examples are possible depending on the application and purpose. For example, the electricity price prediction system 1000 may include multiple prediction devices 10 and one or more terminal devices 20. For example, the prediction device 10 may be realized by multiple computers, or may be realized as a cloud computing service. For example, the electricity price prediction system 1000 may be realized by a standalone computer. The classification of devices such as the prediction device 10 and the terminal device 20 shown in FIG. 1 is one example.

[0030] <Hardware Configuration> The hardware configuration of each device included in the electricity price prediction system 1000 in the first embodiment will be described with reference to FIG.

[0031] <Computer> The prediction device 10 and the terminal device 20 in the first embodiment are realized by, for example, a computer. Fig. 2 is a block diagram showing an example of the hardware configuration of the computer.

[0032] 2, the computer 500 includes a CPU (Central Processing Unit) 501, a ROM (Read Only Memory) 502, a RAM (Random Access Memory) 503, a HDD (Hard Disk Drive) 504, an input device 505, a display device 506, a communication I / F (Interface) 507, and an external I / F 508. The CPU 501, the ROM 502, and the RAM 503 form a so-called computer. The hardware components of the computer 500 are connected to each other via a bus line 509. The input device 505 and the display device 506 may be connected to the external I / F 508 for use.

[0033] The CPU 501 is a computing device that reads programs and data from a storage device such as the ROM 502 or the HDD 504 onto the RAM 503 and executes processing to realize overall control and functions of the computer 500. The computer 500 may have a GPU (Graphics Processing Unit) in addition to or instead of the CPU 501.

[0034] The ROM 502 is an example of a non-volatile semiconductor memory (storage device) that can retain programs and data even when the power is turned off. The ROM 502 functions as a main storage device that stores various programs, data, etc. required for the CPU 501 to execute various programs installed in the HDD 504. Specifically, the ROM 502 stores boot programs such as a Basic Input / Output System (BIOS) and an Extensible Firmware Interface (EFI) that are executed when the computer 500 starts up, as well as data such as OS (Operating System) settings and network settings.

[0035] The RAM 503 is an example of a volatile semiconductor memory (storage device) in which programs and data are erased when the power is turned off. The RAM 503 is, for example, a dynamic random access memory (DRAM) or a static random access memory (SRAM). The RAM 503 provides a working area in which various programs installed in the HDD 504 are expanded when executed by the CPU 501.

[0036] The HDD 504 is an example of a non-volatile storage device that stores programs and data. The programs and data stored in the HDD 504 include an OS, which is basic software that controls the entire computer 500, and applications that provide various functions on the OS. Note that the computer 500 may use a storage device that uses flash memory as a storage medium (e.g., an SSD (Solid State Drive)) instead of the HDD 504.

[0037] The input device 505 includes a touch panel, operation keys and buttons, a keyboard and mouse, a microphone for inputting sound data such as voice, and the like, which are used by the user to input various signals.

[0038] The display device 506 is composed of a display such as a liquid crystal display or organic electroluminescence (EL) display for displaying a screen, a speaker for outputting sound data such as voice, and the like.

[0039] The communication I / F 507 is an interface that connects to a communication network and enables the computer 500 to perform data communication.

[0040] The external I / F 508 is an interface with external devices, such as a drive device 510.

[0041] The drive device 510 is a device for loading a recording medium 511. The recording medium 511 here includes media that record information optically, electrically, or magnetically, such as CD-ROMs, flexible disks, and magneto-optical disks. The recording medium 511 may also include semiconductor memories that record information electrically, such as ROMs and flash memories. This allows the computer 500 to read from and / or write to the recording medium 511 via the external I / F 508.

[0042] The various programs to be installed in the HDD 504 are installed, for example, by setting the distributed recording medium 511 in a drive device 510 connected to the external I / F 508 and reading the various programs recorded on the recording medium 511 by the drive device 510. Alternatively, the various programs to be installed in the HDD 504 may be installed by being downloaded via the communication I / F 507 from a network different from the communication network.

[0043] <Functional Configuration> The functional configuration of the prediction device 10 in the first embodiment will be described with reference to Fig. 3. Fig. 3 is a block diagram showing an example of the functional configuration of the prediction device in the first embodiment.

[0044] As shown in FIG. 3, the prediction device 10 includes an actual data acquisition unit 101, a data storage unit 102, a teacher data generation unit 103, a model learning unit 104, a model storage unit 105, a verification prediction unit 106, an accuracy calculation unit 107, a model selection unit 108, and a target prediction unit 109.

[0045] The performance data acquisition unit 101, the teacher data generation unit 103, the model learning unit 104, the verification prediction unit 106, the accuracy calculation unit 107, the model selection unit 108, and the target prediction unit 109 are realized, for example, by a process in which a program expanded from the HDD 504 shown in FIG. 2 onto the RAM 503 is executed by the CPU 501.

[0046] The data storage unit 102 and the model storage unit 105 are realized using, for example, the RAM 503 or the HDD 504 shown in FIG.

[0047] The performance data acquisition unit 101 acquires performance data from the external data source 30. The performance data acquisition unit 101 acquires performance data at predetermined time intervals. The predetermined time interval may be, for example, within one week.

[0048] The actual data includes electricity market price data and LNG price data. The electricity market price data is time-series data that records actual values ​​of electricity market prices in the wholesale electricity trading market. The LNG price data is time-series data that records actual values ​​of index prices in the LNG market.

[0049] The performance data may include at least one of weather forecast data and exchange rate data. The weather forecast data is time-series data that records past weather forecasts for a target region for which the electricity market price is predicted. The exchange rate data is time-series data that records actual exchange rates in the foreign exchange market.

[0050] The performance data acquisition unit 101 may acquire electricity market price data from the electricity market system 31. The performance data acquisition unit 101 may acquire LNG price data from the LNG market system 32. The performance data acquisition unit 101 may acquire weather forecast data from the weather information system 33. The performance data acquisition unit 101 may acquire exchange rate data from the exchange market system 34. If there is another data source that accumulates and provides any performance data, the performance data acquisition unit 101 may acquire the performance data from that data source.

[0051] The data storage unit 102 stores performance data acquired by the performance data acquisition unit 101. Every time the performance data acquisition unit 101 acquires performance data, the most recently acquired performance data is accumulated in the data storage unit 102.

[0052] The teacher data generation unit 103 generates multiple pieces of teacher data based on the performance data read from the data storage unit 102. The multiple pieces of teacher data have different learning periods. The learning period is the time range during which the performance data included in the teacher data occurred (more specifically, the period from the time when the oldest performance data among the performance data included in the teacher data occurred to the time when the newest performance data occurred).

[0053] The plurality of teacher data may have different learning periods. The plurality of teacher data may have the same learning period but different start and end times. The length of the learning period may be set arbitrarily, and may be, for example, from several months to one year.

[0054] The training data includes some or all of the following data items. However, at least the electricity market price is included in the training data. The electricity market price is used as the objective variable. The other data items are used as explanatory variables. The index price in the LNG market is an example of a fuel price. The fuel price is the price of fuel used in power generation. Specifically, the fuel price may be the price of fuel used at a power plant located in the target area for predicting the electricity market price. - Electricity market price - LNG market index price - Exchange rate - Calendar feature - Weather forecast data - Solar power generation forecast - Electricity demand forecast - Lag feature of electricity market price - Statistics of electricity market price

[0055] For the electricity market price, the actual value of the electricity market price during the learning period can be extracted from the electricity market price data. For the LNG market index price, the actual value of the index price during the learning period can be extracted from the LNG price data. For the exchange rate, the actual value of the exchange rate during the learning period can be extracted from the exchange rate data.

[0056] The calendar feature may include, for example, the day of the week, the time, holiday information, etc. The weather forecast data may include, for example, a 39-hour forecast or a 78-hour forecast of the Meso Scale Model - Grid Point Value (MSM-GPV) provided by the Japan Meteorological Agency.

[0057] For example, the solar power generation forecast may be based on weather forecast data. For example, the power demand forecast may be based on weather forecast data and calendar features. For the solar power generation forecast or the power demand forecast, predicted data may be obtained from an external data source 30.

[0058] The lag feature of the electricity market price may be generated by extracting actual values ​​of the electricity market price during a learning period from the electricity market price data, and shifting the extracted electricity market prices by a predetermined time interval. The statistical quantity of the electricity market price may be, for example, a moving average or a moving standard deviation. The moving average or the moving standard deviation may be generated by extracting actual values ​​of the electricity market price during a learning period from the electricity market price data, and calculating the average or standard deviation for each of a predetermined number of the extracted electricity market prices.

[0059] The model learning unit 104 generates a plurality of prediction models based on the plurality of training data generated by the training data generation unit 103. The model learning unit 104 generates a plurality of prediction models at predetermined time intervals. The predetermined time interval may be, for example, within one week.

[0060] The model learning unit 104 learns each of a plurality of pieces of training data based on each of a plurality of predetermined prediction algorithms, thereby generating a prediction model corresponding to each combination of the training data and the prediction algorithm.

[0061] The prediction algorithm includes at least one of a machine learning model and a statistical model. Examples of the machine learning model include random forest, LightGBM (Light Gradient Boosting Machine), or XGBoost (eXtreme Gradient Boosting). Examples of the statistical model include linear regression or an autoregressive model. Examples of the autoregressive model include an autoregressive integrated moving average (ARIMA) model. The prediction algorithm may include multiple types of machine learning models or multiple types of statistical models.

[0062] The model learning unit 104 may generate multiple prediction models with different explanatory variables for each combination of training data and a prediction algorithm. The explanatory variables may be predetermined to be available depending on the prediction algorithm. In this case, the multiple prediction models include prediction models corresponding to each combination of training data, a prediction algorithm, and explanatory variables.

[0063] The model storage unit 105 stores a plurality of prediction models generated by the model learning unit 104. Each time the model learning unit 104 generates a prediction model, the prediction models stored in the model storage unit 105 are updated with the most recently generated prediction model.

[0064] The verification prediction unit 106 predicts the electricity market price for a predetermined verification period based on each of the multiple prediction models read from the model storage unit 105. The verification period may be a time interval immediately before the prediction target period. The prediction target period is one or more time intervals for predicting the electricity market price. The prediction target period may be, for example, the next day, or several days after the next day. For example, if the prediction target period is the next day, the verification period may be a time interval before the previous day (when the electricity market price on the prediction day has not been made public). The length of the verification period may be set arbitrarily, and may be, for example, several days to about one week.

[0065] The accuracy calculation unit 107 calculates the prediction accuracy of each of the multiple prediction models based on the electricity market price predicted by the verification prediction unit 106. The prediction accuracy may be, for example, the mean absolute error (MAE) between the predicted value of the electricity market price and the actual value of the electricity market price. The actual value of the electricity market price may be the electricity market price included in the training data.

[0066] The model selection unit 108 selects a prediction model with high prediction accuracy from among the multiple prediction models based on the prediction accuracy calculated by the accuracy calculation unit 107. The model selection unit 108 may select the prediction model with the highest prediction accuracy, or may select a predetermined number of prediction models from the top in terms of prediction accuracy. When the prediction accuracy is evaluated using the mean absolute error, the model selection unit 108 may select the prediction model with the smallest mean absolute error, or a predetermined number of prediction models from the smallest mean absolute error.

[0067] The target prediction unit 109 predicts the electricity market price for the prediction period based on the prediction model selected by the model selection unit 108. The target prediction unit 109 may predict the electricity market price by inputting the most recent performance data read from the data storage unit 102 into the prediction model as an explanatory variable.

[0068] When multiple prediction models are selected by the model selection unit 108, the target prediction unit 109 may combine the selected multiple prediction models to predict the electricity market price. For example, the combination of prediction models may be performed by ensemble learning or the like.

[0069] The target prediction unit 109 transmits the prediction result of the electricity market price to the terminal device 20. The target prediction unit 109 may output the prediction result of the electricity market price to the display device 506 included in the prediction device 10.

[0070] <Processing Procedure> A prediction method executed by the prediction device 10 in the first embodiment will be described with reference to Fig. 4. The prediction method is a process for predicting the electricity market price during a prediction target period. Fig. 4 is a flowchart showing an example of the prediction method in the first embodiment.

[0071] In step S1, the performance data acquisition unit 101 of the prediction device 10 acquires performance data from the external data source 30. Next, the performance data acquisition unit 101 stores the performance data in the data storage unit 102. If performance data is already stored in the data storage unit 102, the performance data acquisition unit 101 adds the currently acquired performance data for each type of performance data.

[0072] Specifically, the performance data acquiring unit 101 acquires unacquired electricity market price data from the electricity market system 31. Next, the performance data acquiring unit 101 adds the acquired electricity market price data to the electricity market price data stored in the data storage unit 102.

[0073] Furthermore, the performance data acquiring unit 101 acquires unacquired LNG price data from the LNG market system 32. Next, the performance data acquiring unit 101 adds the acquired LNG price data to the LNG price data stored in the data storage unit 102.

[0074] Furthermore, the performance data acquisition unit 101 acquires unacquired weather forecast data from the weather information system 33. Next, the performance data acquisition unit 101 adds the acquired weather forecast data to the weather forecast data stored in the data storage unit 102.

[0075] Then, the performance data acquiring unit 101 acquires exchange rate data from the exchange market system 34. Next, the performance data acquiring unit 101 adds the acquired exchange rate data to the exchange rate data stored in the data storage unit 102.

[0076] In step S2, the teacher data generation unit 103 of the prediction device 10 reads out the performance data stored in the data storage unit 102. Next, the teacher data generation unit 103 generates multiple pieces of teacher data based on the read performance data. The teacher data generation unit 103 sends the generated multiple pieces of teacher data to the model learning unit 104.

[0077] Specifically, the teacher data generation unit 103 generates multiple learning periods with different start times or end times. Next, the teacher data generation unit 103 extracts performance data generated during each learning period from the read performance data. This generates multiple teacher data with different learning periods.

[0078] Furthermore, the teacher data generation unit 103 may generate multiple teacher data sets with different explanatory variables based on multiple teacher data sets with different learning periods. In this case, the teacher data generation unit 103 may extract explanatory variables included in a predetermined combination of explanatory variables from each of the multiple teacher data sets with different learning periods.

[0079] In step S3, the model learning unit 104 of the prediction device 10 receives multiple pieces of teacher data from the teacher data generation unit 103. Next, the model learning unit 104 learns each of the multiple pieces of teacher data based on each of multiple prediction algorithms. This generates a prediction model corresponding to the combination of each piece of teacher data and each prediction algorithm. The model learning unit 104 updates the prediction models stored in the model storage unit 105 with the multiple generated prediction models.

[0080] The model learning unit 104 may assign identification information for identifying each prediction model to the prediction model. The identification information may be a numerical value or a character string assigned to the prediction model according to a predetermined rule, or may be information indicating a combination of training data, a prediction algorithm, and explanatory variables.

[0081] In step S4, the verification prediction unit 106 of the prediction device 10 reads out a plurality of prediction models stored in the model storage unit 105. Next, the verification prediction unit 106 extracts performance data that occurred during the verification period from the performance data read out in step S2.

[0082] Next, the verification prediction unit 106 inputs explanatory variables including the extracted performance data into each of a plurality of prediction models. Each prediction model predicts the electricity market price based on the input explanatory variables and outputs the predicted value. Then, the verification prediction unit 106 sends the predicted value of the electricity market price output from each prediction model to the accuracy calculation unit 107.

[0083] In step S5, the accuracy calculation unit 107 of the prediction device 10 receives the predicted value of the electricity market price from the verification prediction unit 106. Next, the accuracy calculation unit 107 calculates the prediction accuracy for each of the multiple prediction models based on the predicted value of the electricity market price and the actual value of the electricity market price. Specifically, the accuracy calculation unit 107 calculates the mean absolute error between the predicted value of the electricity market price and the actual value of the electricity market price. Then, the accuracy calculation unit 107 sends the prediction accuracy of each prediction model to the model selection unit 108.

[0084] In step S6, the model selection unit 108 of the prediction device 10 receives the prediction accuracy of each prediction model from the accuracy calculation unit 107. Next, the model selection unit 108 selects a prediction model with high prediction accuracy based on the prediction accuracy of each prediction model. Specifically, the model selection unit 108 selects one or more prediction models with smallest mean absolute error. Then, the model selection unit 108 sends information indicating the selected prediction model (hereinafter also referred to as "selection information") to the target prediction unit 109. The selection information may be, for example, identification information assigned to each prediction model.

[0085] In step S7, the target prediction unit 109 of the prediction device 10 receives selection information from the model selection unit 108. Next, the target prediction unit 109 reads out the prediction model indicated by the selection information from the model storage unit 105. If there are multiple prediction models indicated by the selection information, the target prediction unit 109 combines the read out prediction models to generate one prediction model.

[0086] Next, the target prediction unit 109 extracts the most recently generated performance data from the performance data read in step S2. Next, the target prediction unit 109 inputs explanatory variables including the extracted performance data into the prediction model selected in step S6. The prediction model predicts the electricity market price based on the input explanatory variables and outputs the predicted value. Then, the target prediction unit 109 obtains the predicted value of the electricity market price output from the prediction model.

[0087] In step S8, the target prediction unit 109 of the prediction device 10 transmits the prediction result of the electricity market price to the terminal device 20. The prediction result of the electricity market price includes the predicted value of the electricity market price acquired in step S7. The prediction result of the electricity market price may also include predicted values ​​of the electricity market price predicted in the past in chronological order.

[0088] The terminal device 20 receives the prediction result of the electricity market price from the prediction device 10. Next, the terminal device 20 presents the prediction result of the electricity market price to the user. For example, the terminal device 20 may display the prediction result of the electricity market price on the display device 506. The terminal device 20 may display the prediction result of the electricity market price in any display format. For example, the terminal device 20 may display the transition of the electricity market price in a graph.

[0089] The prediction device 10 repeatedly executes the processes from step S1 to step S6 and the processes from step S7 to step S8 at predetermined time intervals. The predetermined time interval may be set arbitrarily and may be, for example, within one week. The time interval for executing the processes from step S1 to step S6 and the time interval for executing the processes from step S7 to step S8 may be the same as or different from each other. In other words, the prediction device 10 repeatedly executes the generation of a prediction model and the selection of a prediction model at predetermined time intervals. As a result, the prediction device 10 can predict the electricity market price based on a prediction model with high prediction accuracy among prediction models updated based on the most recent performance data.

[0090] According to one aspect, the first embodiment can accurately predict the electricity market price.

[0091] The multiple prediction models may differ from each other in the periods in which the training data was generated. The multiple prediction models may differ from each other in prediction algorithms. The multiple prediction models may include at least one of a machine learning model or a statistical model. The multiple prediction models may differ from each other in explanatory variables. The explanatory variables may include an index price in the LNG market. Therefore, according to the first embodiment, it is possible to select a prediction model with high prediction accuracy from multiple prediction models that have different training data, prediction algorithms, or explanatory variables.

[0092] The verification period may be the time interval immediately preceding the target prediction period. Because the electricity market price is determined by a complex combination of various factors, it is believed that the most recent performance data can be used to select the optimal prediction model at that time. Therefore, according to this embodiment, an appropriate prediction model can be selected depending on the target prediction period.

[0093] The prediction device 10 may repeatedly generate a prediction model and select a prediction model at predetermined time intervals. Because factors affecting the electricity market price change from moment to moment, updating the prediction model at short intervals is considered to improve prediction accuracy. Therefore, according to this embodiment, the prediction model can be appropriately updated based on the most recent performance data.

[0094] The prediction device 10 may select multiple prediction models with high prediction accuracy and combine the selected multiple prediction models to predict the electricity market price. For example, if there are multiple prediction models with high prediction accuracy, combining the prediction results of these models may further improve the prediction accuracy. Therefore, according to this embodiment, multiple prediction models can be combined to accurately predict the electricity price.

[0095] Second Embodiment A second embodiment of the present invention is an electricity price prediction system that predicts electricity prices. The electricity price prediction system in the second embodiment has a function of predicting electricity prices using the prices of fuel used for power generation. In this embodiment, as in the first embodiment, a configuration for predicting an electricity market price as an example of an electricity price will be described.

[0096] In the first embodiment, a configuration for predicting an electricity market price using the most recent fuel price is described. In the second embodiment, a configuration for predicting an electricity market price using past fuel prices is described. In this embodiment, a configuration for using an index price in the LNG market as an example of the fuel price is described. However, the fuel price may be the price of another fuel used for power generation.

[0097] The electricity price forecasting system according to the second embodiment will be described below, focusing on the differences from the first embodiment.

[0098] <Functional Configuration> The functional configuration of the prediction device 10 in the second embodiment will be described with reference to Fig. 5. Fig. 5 is a block diagram showing an example of the functional configuration of the prediction device in the second embodiment.

[0099] 5 , the prediction device 10 includes a performance data acquisition unit 101, a data storage unit 102, a teacher data generation unit 103, a model learning unit 104, a model storage unit 105, a target prediction unit 109, and a time length determination unit 110. That is, the prediction device 10 in the second embodiment differs from the first embodiment in that it does not include the verification prediction unit 106, the accuracy calculation unit 107, and the model selection unit 108, but further includes the time length determination unit 110.

[0100] The teacher data generation unit 103 in this embodiment generates teacher data based on performance data read from the data storage unit 102. In other words, in this embodiment, there may be only one piece of teacher data. The data items included in the teacher data are the same as those in the first embodiment.

[0101] The model learning unit 104 in this embodiment generates a prediction model based on the training data generated by the training data generating unit 103. In other words, in this embodiment, there may be only one prediction model. The prediction algorithm and explanatory variables may be determined in advance. In this embodiment, the explanatory variables of the prediction model include at least the index price in the LNG market.

[0102] The time length determination unit 110 determines the time length n to be used for prediction based on the electricity market price and the LNG market index price included in the training data generated by the training data generation unit 103. Specifically, the time length determination unit 110 first calculates, for each of a plurality of different time lengths n, the price at the point in time going back in time n (the price n days ago) or the moving average price for the time length n (the n-day moving average price). Next, the time length determination unit 110 calculates, for each of a plurality of different time lengths n, a correlation coefficient between the trend in the actual value of the electricity market price and the price n days ago or the trend in the n-day moving average price. Then, the time length determination unit 110 determines the time length n with the highest correlation coefficient as the time length n to be used for prediction.

[0103] The target prediction unit 109 in this embodiment predicts the electricity market price for the prediction target period using the time length n determined by the time length determination unit 110. The target prediction unit 109 may read the index price of the LNG market from the data storage unit 102 and obtain the price n days ago. Alternatively, the target prediction unit 109 may read the index price of the LNG market from the data storage unit 102 and calculate an n-day moving average. The target prediction unit 109 may predict the electricity market price by inputting explanatory variables including the price n days ago or the n-day moving average into the prediction model read from the model storage unit 105.

[0104] <Processing Procedure> A prediction method executed by the prediction device 10 in the second embodiment will be described with reference to Fig. 6. Fig. 6 is a flowchart showing an example of the prediction method in the second embodiment.

[0105] In step S11, the performance data acquisition unit 101 of the prediction device 10 acquires performance data from the external data source 30. Next, the performance data acquisition unit 101 stores the performance data in the data storage unit 102.

[0106] In step S12, the teacher data generation unit 103 of the prediction device 10 reads out the performance data stored in the data storage unit 102. Next, the teacher data generation unit 103 generates teacher data based on the read performance data. Specifically, the teacher data generation unit 103 extracts performance data generated during a predetermined learning period. The teacher data generation unit 103 sends the generated teacher data to the model learning unit 104 and the time length determination unit 110.

[0107] In step S13, the model learning unit 104 of the prediction device 10 receives the teacher data from the teacher data generation unit 103. Next, the model learning unit 104 learns the teacher data based on a predetermined prediction algorithm. The model learning unit 104 updates the prediction model stored in the model storage unit 105 with the generated prediction model.

[0108] In step S14, the time length determination unit 110 of the prediction device 10 receives the training data from the training data generation unit 103. Next, the time length determination unit 110 obtains the index price of the LNG market included in the training data, and calculates the price n days ago or the n-day moving average price for each of a plurality of different time lengths n.

[0109] Next, the time length determination unit 110 calculates a correlation coefficient between the trend in the actual value of the electricity market price and the trend in the price n days ago or the n-day moving average price for each of multiple different time lengths n. Next, the time length determination unit 110 determines the time length n with the highest correlation coefficient as the time length n to be used for prediction. Then, the time length determination unit 110 sends the determined time length n to the target prediction unit 109.

[0110] In step S15, the target prediction unit 109 of the prediction device 10 receives the duration n from the duration determination unit 110. Next, the target prediction unit 109 reads out a prediction model from the model storage unit 105.

[0111] Next, the target prediction unit 109 extracts the most recently generated actual data from the actual data read in step S12. However, for the index price in the LNG market, the target prediction unit 109 calculates the price n days ago or the n-day moving average using the time length n determined in step S14. Next, the target prediction unit 109 inputs explanatory variables including the price n days ago or the n-day moving average into the read prediction model. The prediction model predicts the electricity market price based on the input explanatory variables and outputs the predicted value. Then, the target prediction unit 109 obtains the predicted value of the electricity market price output from the prediction model.

[0112] In step S16, the target prediction unit 109 of the prediction device 10 transmits the prediction result of the electricity market price to the terminal device 20. The terminal device 20 receives the prediction result of the electricity market price from the prediction device 10. Next, the terminal device 20 presents the prediction result of the electricity market price to the user.

[0113] The prediction device 10 repeatedly executes the processes from step S11 to step S14 and the processes from step S15 to step S16 at predetermined time intervals. The predetermined time interval may be set arbitrarily and may be, for example, within one week. The time interval for executing the processes from step S11 to step S14 and the time interval for executing the processes from step S15 to step S16 may be the same as or different from each other. In other words, the prediction device 10 repeatedly executes generating a prediction model and determining the time length n at predetermined time intervals. This allows the prediction device 10 to predict the electricity market price based on the prediction model and the time length updated based on the most recent performance data.

[0114] <Prediction Accuracy> The prediction accuracy of the prediction device 10 in the second embodiment will be described with reference to FIGS. 7 and 8. FIG.

[0115] FIG. 7 is a scatter plot showing an example of the prediction accuracy for each index price. In FIG. 7, the vertical axis represents the predicted value of the electricity market price, and the horizontal axis represents the actual value of the electricity market price, and the prediction results are plotted. In the scatter plot shown in FIG. 7, the dark plots represent the prediction results using the LNG price seven days in advance (the price n days in advance when n = 7). The light plots represent the prediction results using the LNG price on the current day. As shown in FIG. 7, the prediction results using the LNG price seven days in advance have smaller variance and higher prediction accuracy than the prediction results using the LNG price on the current day.

[0116] Figure 8 is a line graph showing an example of prediction accuracy for each index price. In Figure 8, the vertical axis represents the electricity market price and the horizontal axis represents time, and the graph shows trends in the actual value of the electricity market price, the predicted value using the LNG price on the day, and the predicted value using the LNG price seven days ago. In the line graph shown in Figure 8, the dashed line represents the actual value, the thick solid line represents the predicted value using the LNG price seven days ago, and the thin solid line represents the predicted value using the LNG price on the day. As shown in Figure 8, it can be seen that the predicted value using the LNG price seven days ago has a smaller error from the actual value and higher prediction accuracy than the predicted value using the LNG price on the day.

[0117] Table 1 shows the results of calculating the mean absolute error for the forecast value using the LNG price of the day and the forecast value using the LNG price seven days in advance. As shown in Table 1, the forecast value using the LNG price of the day has a smaller mean absolute error than the forecast value using the LNG price seven days in advance, indicating higher prediction accuracy.

[0118]

[0119] As described above, Figures 7, 8 and Table 1 show that when predicting the electricity market price using the index price in the LNG market as an explanatory variable, the electricity market price can be predicted with higher accuracy by using the index price from several days prior rather than the index price on the current day.

[0120] The prediction device 10 according to the second embodiment predicts the electricity market price by acquiring the fuel price indicating the price of fuel used for power generation and inputting explanatory variables including the fuel price into a trained prediction model. According to one aspect, the second embodiment makes it possible to predict the electricity price with high accuracy.

[0121] The fuel price may be an index price in the LNG market. The fuel price may be a price at a time point a predetermined length of time prior to the forecast period for predicting the electricity market price. The fuel price may be a moving average price for a predetermined length of time prior to the forecast period for predicting the electricity market price. It has been found that electricity market prices can be predicted more accurately by using past fuel prices rather than current fuel prices. Therefore, according to the second embodiment, electricity prices can be predicted with high accuracy.

[0122] The prediction device 10 may determine the time length based on a correlation coefficient between the change in the electricity market price and the change in the fuel price. It is considered that the electricity market price can be predicted more accurately by using fuel prices from a period when the correlation between the electricity market price and the fuel price is high among past fuel prices. Therefore, according to the second embodiment, the electricity price can be predicted with high accuracy.

[0123] The prediction device 10 may repeatedly generate a prediction model and determine the time length at predetermined time intervals. Because factors affecting the electricity market price change from moment to moment, updating the time length at short intervals is considered to improve prediction accuracy. Therefore, according to this embodiment, the time length can be appropriately updated based on the most recent performance data.

[0124] [Supplementary Note] Each function of the above-described embodiments can be realized by one or more processing circuits. Here, the term "processing circuit" in this specification includes a processor programmed to execute each function by software, such as a CPU (Central Processing Unit) or GPU (Graphics Processing Unit) implemented by an electronic circuit, as well as devices such as an ASIC (Application Specific Integrated Circuit), DSP (Digital Signal Processor), FPGA (Field Programmable Gate Array), and conventional circuit modules designed to execute each of the above-described functions.

[0125] The disclosed technology may take the following forms: (Supplementary Note 1) A prediction device comprising: a verification prediction unit that predicts an electricity price for a predetermined verification period based on each of a plurality of trained prediction models; a model selection unit that selects a prediction model from the plurality of prediction models that has a high prediction accuracy for the electricity price; and a target prediction unit that predicts the electricity price for a predetermined target prediction period based on the selected prediction model. (Supplementary Note 2) The prediction device according to Supplementary Note 1, wherein the plurality of prediction models have different time ranges of training data used for learning. (Supplementary Note 3) The prediction device according to Supplementary Note 1, wherein the plurality of prediction models have different prediction algorithms. (Supplementary Note 4) The prediction device according to Supplementary Note 1, wherein the plurality of prediction models have different explanatory variables. (Supplementary Note 5) The prediction device according to Supplementary Note 4, wherein the explanatory variables include a price of fuel used for power generation. (Supplementary Note 6) The prediction device according to Supplementary Note 1, wherein the verification period is a time interval immediately preceding the target prediction period. (Supplementary Note 7) The prediction device according to any one of Supplements 1 to 6, further comprising a model learning unit that generates the plurality of prediction models, and repeatedly performs the steps of generating the prediction models by the model learning unit and selecting the prediction model by the model selection unit at predetermined time intervals. (Supplementary Note 8) The prediction device according to any one of Supplements 1 to 6, wherein the model selection unit selects the plurality of prediction models with high prediction accuracy, and the target prediction unit predicts the electricity price by combining the selected plurality of prediction models. (Supplementary Note 9) A prediction device comprising: an acquisition unit that acquires a fuel price indicating the price of fuel used for power generation; and a prediction unit that predicts the electricity price by inputting explanatory variables including the fuel price into a trained prediction model. (Supplementary Note 10) The prediction device according to Supplementary Note 9, wherein the fuel price is an index price in the liquefied natural gas market. (Supplementary Note 11) The prediction device according to Supplementary Note 9, wherein the fuel price is a price at a point in time that is a predetermined length of time prior to a prediction period for which the electricity price is predicted. (Supplementary Note 12) The prediction device according to Supplementary Note 9, wherein the fuel price is a moving average price for a predetermined time period prior to a prediction target period for predicting the electricity price.(Supplementary Note 13) The prediction device according to Supplementary Note 11 or 12, further comprising a time length determination unit that determines the time length based on a correlation coefficient between a change in the electricity price and a change in the fuel price. (Supplementary Note 14) The prediction device according to Supplementary Note 13, further comprising a model learning unit that generates the prediction model, and repeatedly performs the steps of generating the prediction model by the model learning unit and determining the time length by the time length determination unit at predetermined time intervals. (Supplementary Note 15) A prediction method, wherein a computer executes the steps of: predicting an electricity price for a predetermined verification period based on each of a plurality of learned prediction models; selecting a prediction model from the plurality of prediction models that has a high prediction accuracy for the electricity price; and predicting the electricity price for a predetermined prediction period based on the selected prediction model.

[0126] Although the embodiments of the present invention have been described in detail above, the present invention is not limited to these embodiments, and various modifications and changes are possible within the scope of the gist of the present invention described in the claims.

[0127] This application claims priority from Japanese Patent Application No. 2023-175622, filed on October 11, 2023 with the Japan Patent Office, the entire contents of which are incorporated herein by reference.

[0128] 10: Prediction device 20: Terminal device 31: Electricity market system 32: LNG market system 33: Weather information system 34: Foreign exchange market system 101: Performance data acquisition unit 102: Data storage unit 103: Teacher data generation unit 104: Model learning unit 105: Model storage unit 106: Verification prediction unit 107: Accuracy calculation unit 108: Model selection unit 109: Target prediction unit 110: Time length determination unit 1000: Electricity price prediction system

Claims

1. A prediction device comprising: a verification prediction unit that predicts electricity prices during a specified verification period based on each of a plurality of trained prediction models; a model selection unit that selects a prediction model from the plurality of prediction models that has high prediction accuracy for the electricity price; and a target prediction unit that predicts the electricity price during a specified prediction period based on the selected prediction model.

2. The prediction device according to claim 1, wherein the plurality of prediction models have different time ranges of training data used for learning.

3. The prediction device according to claim 1, wherein the plurality of prediction models have different prediction algorithms.

4. The prediction device according to claim 1, wherein the multiple prediction models have mutually different explanatory variables.

5. The prediction device according to claim 4, wherein the explanatory variables include a price of fuel used in power generation.

6. The prediction device according to claim 1, wherein the validation period is a time interval immediately preceding the prediction target period.

7. A prediction device as described in any one of claims 1 to 6, further comprising a model learning unit that generates the plurality of prediction models, wherein the generating of the prediction models by the model learning unit and the selecting of the prediction model by the model selection unit are repeatedly performed at a predetermined time interval.

8. A prediction device as described in any one of claims 1 to 6, wherein the model selection unit selects a plurality of the prediction models having high prediction accuracy, and the target prediction unit predicts the electricity price by combining the selected plurality of the prediction models.

9. A prediction device comprising: an acquisition unit that acquires a fuel price indicating the price of fuel used for power generation; and a prediction unit that predicts electricity prices by inputting explanatory variables including the fuel price into a trained prediction model.

10. The forecasting device according to claim 9, wherein the fuel price is an index price in a liquefied natural gas market.

11. The prediction device according to claim 9, wherein the fuel price is a price at a point in time going back a predetermined length of time from a forecast period for which the electricity price is predicted.

12. The prediction device according to claim 9, wherein the fuel price is a moving average price for a predetermined time period prior to a forecast period for predicting the electricity price.

13. The prediction device according to claim 11 or 12, further comprising a time length determination unit that determines the time length based on a correlation coefficient between the change in the electricity price and the change in the fuel price.

14. The prediction device according to claim 13, further comprising a model learning unit that generates the prediction model, wherein the generating of the prediction model by the model learning unit and the determining of the time length by the time length determination unit are repeatedly performed at a predetermined time interval.

15. A prediction method in which a computer executes the steps of: predicting electricity prices for a specified verification period based on each of a plurality of learned prediction models; selecting a prediction model from the plurality of prediction models that has high prediction accuracy for the electricity price; and predicting the electricity price for a specified prediction period based on the selected prediction model.

Images