Electricity contract price prediction device and electricity contract price prediction method
The power contract price prediction device uses temperature-dependent electricity usage and solar power generation data to enhance prediction accuracy and stability, addressing the limitations of existing methods by incorporating machine learning for precise contract price forecasting.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- NAT UNIV CORP KYUSHU INST OF TECH (JP)
- Filing Date
- 2022-03-15
- Publication Date
- 2026-05-15
Smart Images
Figure 0007859652000001 
Figure 0007859652000002 
Figure 0007859652000003
Abstract
Description
Technical Field
[0001] The present invention relates to a power contract price prediction device and a power contract price prediction method for predicting the contract price of power trading in the Japan Electricity Exchange.
Background Art
[0002] In the spot market (one-day-ahead market), which is one of the main power trading markets in the Japan Electricity Exchange (JEPX), the contract price of power trading is determined for each of the 48 segments obtained by dividing a day into 30-minute intervals. Buyers and sellers submit bids by specifying the desired transaction price (the desired bid price for buyers and the desired ask price for sellers) and the desired transaction volume by 10:00 am, and the price at the intersection of the demand curve based on the buyers' bids and the supply curve based on the sellers' bids is determined as the contract price.
[0003] A power trading transaction at the contract price is concluded between buyers who bid at prices higher than the contract price and sellers who bid at prices lower than the contract price, and buyers who bid at prices lower than the contract price and sellers who bid at prices higher than the contract price do not conclude a transaction. Note that the unit of the contract price and the desired transaction price for bidding is yen / kWh. Here, since bids are submitted for each segment and the contract price is determined, if the contract price can be predicted, the buyer can select segments with a lower contract price and submit bids, thereby suppressing the power procurement cost. Therefore, a system that can predict the contract price is required, and specific examples thereof are disclosed in, for example, Patent Documents 1 and 2.
Prior Art Documents
Patent Documents
[0004]
Patent Document 1
Patent Document 2
Summary of the Invention
Problems to be Solved by the Invention
[0005] The more accurate the prediction of the execution price, the better, and there is a need for a system that can predict execution prices more stably and accurately than conventional methods. This invention has been made in view of the above circumstances, and aims to provide a power contract price prediction device and a power contract price prediction method that can stably predict the contract price in the wholesale power market. [Means for solving the problem]
[0006] The first electricity contract price prediction device, which is in line with the above objective, is an electricity contract price prediction device that predicts the contract price of electricity buying and selling in a target area determined in the wholesale electricity market for each of several time slots into which a day is divided into predetermined unit time intervals, and comprises a calculation means that uses either or both of the following as explanatory variables: a temperature-dependent predicted amount of electricity usage derived by weighting either the population of each district in the target area and the predicted temperature of each district in the target area with the other, and a predicted amount of solar power generation obtained by weighting either the amount of solar power generation installed in each district of the target area and the predicted amount of solar radiation of each district in the target area, obtained from the status of installation of solar power generation equipment, with the objective variable being the predicted value of the contract price, which is calculated by machine learning.
[0007] A second invention relating to the above-mentioned objective is a method for predicting electricity contract prices, which predicts the contract price for buying and selling electricity in a target area determined in the wholesale electricity market for each of several time slots into which a day is divided into predetermined unit time intervals. The method uses either or both of the following as explanatory variables: a temperature-dependent predicted amount of electricity usage derived by weighting either the population of each district in the target area or the predicted temperature of each district in the target area with the other, and a predicted amount of solar power generation obtained by weighting either the amount of solar power generation installed in each district of the target area or the predicted amount of solar radiation of each district in the target area, obtained from the status of solar power generation installations, with the predicted value of the contract price, which is the objective variable, to be determined by machine learning. [Effects of the Invention]
[0008] The electricity contract price prediction device according to the first invention and the electricity contract price prediction method according to the second invention use either or both of the following as explanatory variables: a temperature-dependent predicted amount of electricity usage derived by weighting either the population of each district in the target area and the predicted temperature of each district in the target area with the other, and a predicted amount of solar power generation obtained by weighting either the amount of solar power generation installed in each district in the target area and the predicted amount of solar radiation in each district in the target area, obtained from the status of solar power generation installation, with the predicted value of the contract price, which is the dependent variable, as determined by machine learning.
[0009] Since it was confirmed that the predicted amount of temperature-dependent electricity usage has a high correlation with the contract price determined in the wholesale electricity market compared to the predicted temperature of the target area, and that the predicted amount of solar power generation has a high correlation with the same contract price compared to the predicted amount of solar radiation of the target area, the electricity contract price prediction device according to the first invention and the electricity contract price prediction method according to the second invention can reliably predict the contract price in the wholesale electricity market. [Brief explanation of the drawing]
[0010] [Figure 1] This is an explanatory diagram showing how the electricity contract price prediction device according to one embodiment of the present invention predicts the contract price. [Figure 2] This is an explanatory diagram showing the locations where predicted temperatures are predicted, obtained by the temperature prediction acquisition method. [Figure 3] This is an explanatory diagram of the area data of predicted temperature, which was generated from point data of predicted temperature. [Figure 4] This is an explanatory diagram showing how the algorithm of the calculation means is updated. [Figure 5] This is an explanatory diagram showing how the minimum price expected value derivation mechanism retrieves information from the performance storage unit. [Figure 6] (A) is an explanatory diagram showing the correlation between the highest predicted temperature-dependent electricity usage and the highest actual transaction price, and (B) is an explanatory diagram showing the correlation between the predicted maximum temperature and the highest actual transaction price. [Figure 7](A) is an explanatory diagram showing the correlation between the highest predicted solar power generation amount and the lowest actual transaction price, and (B) is an explanatory diagram showing the correlation between the highest predicted solar radiation amount and the lowest actual transaction price. [Figure 8] This is an explanatory diagram showing experimental results comparing predicted and actual transaction prices. [Modes for carrying out the invention]
[0011] Next, with reference to the attached drawings, embodiments of the present invention will be described to facilitate understanding of the invention. As shown in Figure 1, the electricity contract price prediction device 10 according to one embodiment of the present invention is a device that predicts the contract price (hereinafter also simply referred to as "contract price") for electricity buying and selling in a target area, which is determined in the wholesale electricity market, for each of several time slots into which the day is divided into predetermined unit time intervals, and includes a calculation means 11 that calculates the predicted value of the contract price by machine learning. A detailed explanation follows below.
[0012] In this embodiment, the target area for predicting contract prices is the Kyushu area, the unit time for one session of the wholesale electricity market is 30 minutes, and the number of sessions per day is 48. However, it goes without saying that the target area, unit time, and number of sessions per day are not limited to these. In this embodiment, the electricity contract price prediction device 10 predicts the contract price for each of the 48 sessions of the prediction target day (hereinafter also referred to as the "prediction target day") on the day before the bidding target day of the wholesale electricity market. Users (buyers) of the electricity contract price prediction device 10 can decide which session and price to bid on by a predetermined time on the day before, by referring to the predicted contract price of the electricity contract price prediction device 10.
[0013] The electricity contract price prediction device 10 can be configured as a computer capable of acquiring electronic data from an external source via an internet connection or the like. The computer is equipped with a storage medium for storing software programs and various electronic data, a CPU, input devices, and the like. As shown in Figure 1, the electricity contract price prediction device 10 includes a predicted temperature acquisition means 12 that obtains the predicted temperature for the target day from an external source, and a population storage unit 13 that stores the population of each district in the target area.
[0014] What the predicted temperature acquisition means 12 obtains from the outside is the predicted temperature every hour (not necessarily every hour) of the prediction target day for each of a plurality of points within the target area (in this embodiment, the predicted temperature of GPV provided by the Japan Meteorological Agency). As shown by the points (◆) in FIG. 2, the latitude and longitude of each point are determined by an external agency so that the points are provided at equal intervals within the target area. The Japan Meteorological Agency provides the predicted temperature every hour at each point for a plurality of times a day, such as the next day or the day after tomorrow. The predicted temperature acquisition means 12 acquires (not limited to once a day) the information of the predicted temperature every hour (24 in total) of the prediction target day at a determined time.
[0015] On the other hand, what the population storage unit 13 stores is the population for each municipality (an example of an area) in the target area (the daytime population or the nighttime population may be adopted), and the target area of the predicted temperature obtained by the predicted temperature acquisition means 12 from the outside does not match the target area of the population stored in the population storage unit 13. Therefore, in this embodiment, as shown in FIG. 1, a predicted temperature conversion means 14 is provided in the electricity contract price prediction device 10, and the predicted temperature of each municipality in the target area is derived from the predicted temperature of each point in the target area obtained by the predicted temperature acquisition means 12 from the outside using a geographic information system (GIS).
[0016] Specifically, the predicted temperature conversion means 14 acquires the predicted temperature of each point at the time of a predetermined frame (for example, the frame at 13:00) of the prediction target day obtained by the predicted temperature acquisition means 12 from the outside, and inputs each predicted temperature as point data to the corresponding position on the map data. Next, interpolation calculation is performed based on all the point data of the predicted temperature input to the map data to generate surface data of the predicted temperature as shown in FIG. 3, and the predicted temperature of a predetermined frame of the prediction target day for each municipality in the target area is derived using the surface data of the predicted temperature. The predicted temperature conversion means 14 performs this process for each of the 48 frames of the prediction target day, and obtains 48 predicted temperatures for each municipality.
[0017] As shown in FIG. 1, the electricity contract price prediction device 10 includes a power consumption prediction means 15 that acquires the population of each municipality in the target area and the predicted temperature of each municipality in the target area from a population storage unit 13 and a predicted temperature conversion means 14, respectively. The power consumption prediction means 15 weights either the population of each municipality in the target area acquired from the population storage unit 13 or the predicted temperature of each time slot of the predicted target day for each municipality in the target area acquired from the predicted temperature conversion means 14 with the other, and derives a temperature-dependent power consumption prediction amount.
[0018] In the present embodiment, the following formula 1 is adopted as the derivation formula of the temperature-dependent power consumption prediction amount. Temperature-dependent power consumption prediction amount [K] = {Σ (population of each area [persons] × predicted temperature of each area [K])} ÷ total population of the target area [persons] ··· Formula 1
[0019] The temperature-dependent power consumption prediction amount is a predicted value of the power consumption by electric devices such as air conditioners that are greatly affected by temperature, and is a value that increases or decreases by weighting the population of each area and the temperature of each area. In view of the fact that the power consumption by the use of the electric device greatly contributes to the contract price, the temperature-dependent power consumption prediction amount is used as an explanatory variable for the predicted value of the contract price, which is the target variable. Verification has confirmed that the accuracy of calculating the predicted value of the contract price is improved by using the temperature-dependent power consumption prediction amount as an explanatory variable.
[0020] The derivation formula of the temperature-dependent power consumption prediction amount is not limited to Formula 1. For example, a coefficient may be calculated by an arithmetic formula with the population of each area as a variable, and the predicted temperature of each area may be weighted with the coefficient (weight) to derive the temperature-dependent power consumption prediction amount, or a coefficient may be calculated by an arithmetic formula with the predicted temperature of each area as a variable, and the population of each area may be weighted with the coefficient (weight) to derive the temperature-dependent power consumption prediction amount.
[0021] Furthermore, the electricity contract price prediction device 10 includes a solar radiation prediction amount acquisition means 16 that obtains the solar radiation prediction amount for the day to be predicted from an external source, and a solar power generation amount storage unit 17 that stores the amount of solar power generation installed in each district of the target area, obtained from the status of installation of solar power generation equipment. The solar radiation prediction amount acquisition means 16 obtains the solar radiation prediction amount (in this embodiment, the solar radiation prediction amount from the GPV provided by the Japan Meteorological Agency) from an external source once a day (not limited to once a day) at a predetermined time for each of the predetermined number of points within the target area, every hour (or not every hour) for the day to be predicted. Similar to the predicted temperature, the solar radiation prediction amount is also provided by the Japan Meteorological Agency multiple times a day, with 24-hour solar radiation prediction amounts for each point for the following day, the day after that, etc. The latitude and longitude of each point are determined by an external organization so that the points are set up at equal intervals within the target area.
[0022] Here, the solar power generation introduction amount storage unit 17 stores the solar power generation introduction amount for each municipality (an example of a district) in the target area, and the target area for the solar radiation prediction amount obtained from an external source by the solar radiation prediction amount acquisition means 16 does not match the target area for the solar power generation introduction amount stored in the solar power generation introduction amount storage unit 17. Therefore, in this embodiment, as shown in Figure 1, the electricity contract price prediction device 10 is provided with a solar radiation prediction amount conversion means 18, and the solar radiation prediction amount for the target day in each municipality (unit of district in which the solar power generation introduction amount storage unit 17 stores the solar power generation introduction amount) in the target area is derived from the solar radiation prediction amount for each point in the target area obtained from an external source by the solar radiation prediction amount acquisition means 16, using a geographic information system.
[0023] Specifically, the solar radiation prediction conversion means 18 obtains the solar radiation prediction amount for each point at a predetermined time frame of the target day obtained from the solar radiation prediction acquisition means 16 from an external source, and inputs each solar radiation prediction amount as point data at the corresponding position on the map data. Next, it performs interpolation calculations based on all the point data of solar radiation prediction amounts input into the map data to generate surface data of solar radiation prediction amounts, and uses the surface data of solar radiation prediction amounts to derive the solar radiation prediction amount for a predetermined time frame of the target day for each municipality in the target area.
[0024] The solar radiation prediction conversion means 18 performs this process for each of the 48 timeframes of the prediction target day, and obtains 48 solar radiation prediction amounts for each municipality. In this embodiment, the inverse distance weighting method is used for the interpolation calculation by the predicted temperature conversion means 14 and the solar radiation prediction conversion means 18, but it is not limited to this, and kriging may be used, for example.
[0025] The electricity contract price prediction device 10 includes a solar power generation prediction means 19 that acquires the amount of solar power generation installed in each municipality of the target area and the amount of solar radiation predicted for each municipality of the target area from a solar power generation installation amount storage unit 17 and a solar radiation prediction amount conversion means 18, respectively. The solar power generation prediction means 19 calculates the solar power generation prediction by weighting one of the solar power generation installation amount for each municipality of the target area acquired from the solar power generation installation amount storage unit 17 and the solar radiation prediction amount for each frame of the target day for each municipality of the target area acquired from the solar radiation prediction amount conversion means 18 with the other.
[0026] In this embodiment, the following Equation 2 is adopted as the derivation formula for the predicted amount of solar power generation. Predicted solar power generation [kWh] = Σ(Solar power generation capacity installed in each area [kW]) × Predicted solar radiation for each district [kWh / m²] 2 ]÷Standard solar radiation intensity[kW / m 2 ]) ...expression 2
[0027] In Equation 2, the standard solar radiation intensity is 1 [kW / m²]. 2 This is a constant used to convert the predicted amount of solar power generation to kWh. There is a tendency for the transaction price to decrease when the predicted amount of solar power generation is high. Verification has confirmed that the accuracy of the transaction price prediction improves by using the predicted amount of solar power generation, which is obtained by multiplying the amount of solar power generation installed in each area by the predicted amount of solar radiation in each area, as one of the explanatory variables.
[0028] The formula for deriving the predicted amount of solar power generation is not limited to Equation 2. For example, coefficients may be calculated using a formula in which the amount of solar power generation installed in each area is a variable, and the predicted amount of solar radiation in each area may be weighted by these coefficients (weights) to derive the predicted amount of solar power generation. Alternatively, coefficients may be calculated using a formula in which the predicted amount of solar radiation in each area is a variable, and the predicted amount of solar power generation installed in each area may be weighted by these coefficients (weights) to derive the predicted amount of solar power generation.
[0029] In this embodiment, the population memory unit 13 stores the population for each ward, such as Hakata Ward, while the solar power generation introduction amount memory unit 17 stores the solar power generation introduction amount on a city basis, not on a ward basis. Therefore, in this embodiment, the district units of the population stored in the population memory unit 13 and the district units of the solar power generation introduction amount stored in the solar power generation introduction amount memory unit 17 differ in some respects, but they may be completely identical. The same applies to each point for the predicted temperature obtained from the outside by the predicted temperature acquisition means 12 and each point for the predicted solar radiation obtained from the outside by the predicted solar radiation acquisition means 16. In this embodiment, the population stored in the population memory unit 13 is updated once every five years, and the solar power generation introduction amount stored in the solar power generation introduction amount memory unit 17 is updated four times a year.
[0030] Furthermore, the electricity contract price prediction device 10 includes a stopped power generation capacity derivation means 20 for determining the power generation capacity of power generation equipment scheduled to be shut down at the power plant on the target date (hereinafter also referred to as "shut-down power generation capacity"), a calendar storage unit 21 for storing calendar information, and a minimum price expected value derivation means 22 for determining the minimum price expected value, which indicates the likelihood that the contract price will be the lowest price of the contract price set in the wholesale electricity market (for example, 0.01 yen / kWh).
[0031] The power generation equipment at each power plant may be shut down for reasons such as malfunctions or inspections. The shut-down power generation capacity derivation means 20 obtains the power generation equipment scheduled to be shut down at each power plant, along with its shut-down capacity, from an external source on a daily basis, and calculates the shut-down power generation capacity for each time slot of the forecast period. Since the shut-down power generation capacity affects the contract price, in this embodiment, the shut-down power generation capacity is used as one of the explanatory variables.
[0032] The calendar information stored in the calendar memory unit 21 consists of a year, month, and day, a numerical value indicating whether or not it is a public holiday. Since the execution price is affected by the time of year (e.g., early summer) and day of the week (including whether or not the predicted date is a public holiday), the calendar information is also considered one of the explanatory variables.
[0033] An analysis of past transaction prices revealed that the tendency for the lowest transaction price to occur is higher in spring and autumn than in summer and winter, and particularly high in April and May. Furthermore, this trend was observed to be higher on holidays and Sundays compared to weekdays and Saturdays, and higher during the day compared to nighttime, with the highest price occurring at 12:30 PM.
[0034] Based on these analysis results, the minimum price expected value derivation method 22 uses the following equation 3 to calculate the minimum price expected value for each time slot on the target day. The expected minimum price for the kth day = predicted solar power generation for the kth frame of the forecast day × coefficient for the month × Coefficient of the day of the week × Predicted solar power generation amount for the kth frame of the target day and the kth frame of the same month in the past A coefficient obtained by comparing the predicted amount of solar power generation for each frame with the average value...Equation 3
[0035] However, in Equation 3, k is an integer from 1 to 48, and the minimum price expected value derivation means 22 sequentially calculates the minimum price expected value for 48 frames from the 1st frame to the 48th frame using Equation 3. The "month coefficient" and "day of the week coefficient" are coefficients predetermined from past transaction price performance, etc., and are larger the more likely the transaction price is to be the minimum price, while the "day of the week coefficient" is larger on holidays than on non-holiday days, even on the same day of the week.
[0036] The coefficient obtained by comparing the predicted solar power generation amount for the kth time slot on the target date with the average value of the predicted solar power generation amount for the kth time slot in the same month in past years is greater when the predicted solar power generation amount for the kth time slot on the target date is greater than or equal to the average value of the predicted solar power generation amount for the kth time slot in the same month over the past X years (for example, the past 5 years), and is greater when the predicted solar power generation amount for the kth time slot on the target date is less than the average value of the predicted solar power generation amount for the kth time slot in the same month over the past X years.
[0037] The minimum price expectation is a variable that acts to bring the predicted value of the contract price closer to or equal to the lowest contract price, depending on the magnitude of the predicted amount of solar power generation derived by the solar power generation prediction means 19 relative to the predicted amount of solar power generation for the same month and time slot derived in the past (the larger the predicted amount of solar power generation relative to the predicted amount of solar power generation for the same month and time slot derived in the past, the closer the predicted value of the contract price will be to or equal to the lowest contract price). When the expected minimum price is included as one of the explanatory variables, it has been confirmed that the predicted value of the execution price becomes more consistently close to the minimum price when the actual execution price is the minimum price, compared to when the expected minimum price is not included as an explanatory variable. Therefore, in this embodiment, the expected minimum price is also included as one of the explanatory variables.
[0038] The calculation means 11 obtains the temperature-dependent predicted power consumption for the target day from the power consumption prediction means 15, the predicted solar power generation for the target day from the solar power generation prediction means 19, the power generation capacity of the power generation equipment scheduled to be shut down at the power plant on the target day from the shutdown power generation capacity derivation means 20, the numerical value of the calendar information for the target day from the calendar storage unit 21, and the expected minimum price for the target day from the minimum price expectation derivation means 22. Using all of these numerical values as explanatory variables, the calculation means 11 calculates the predicted value of the transaction price for each of the 48 time slots of the target day, which is the objective variable, using machine learning. The type of machine learning is not particularly limited; for example, random forests or support vector machines can be used.
[0039] As shown in Figures 1 and 4, the electricity contract price prediction device 10 includes a performance storage unit 23 that obtains and stores explanatory variables used to calculate the predicted contract price from the electricity usage prediction means 15, the solar power generation amount prediction means 19, the stopped power generation capacity derivation means 20, the calendar storage unit 21, and the minimum price expected value derivation means 22, and an update means 24 that updates the calculation means 11 using the numerical values stored in the performance storage unit 23.
[0040] As shown in Figure 1, after the calculation means 11 calculates the predicted value of the contract price for each time slot of the forecast day, the actuals storage unit 23 acquires and stores the temperature-dependent predicted power consumption for each time slot of the forecast day, the predicted solar power generation for each time slot of the forecast day, the power generation capacity of the power generation equipment scheduled to be shut down at the power plant on the forecast day, the numerical value of the calendar information for the forecast day, and the minimum price expected value for each time slot of the forecast day from the power consumption prediction means 15, the solar power generation prediction means 19, the shutdown power generation capacity derivation means 20, the calendar storage unit 21, and the minimum price expected value derivation means 22, respectively. After the contract prices for all 48 time slots of the forecast day have been determined in the wholesale power market, the actual values of the contract prices for each time slot of the forecast day that have been determined are acquired from an external source and stored.
[0041] As shown in Figure 4, the update means 24, after the performance storage unit 23 obtains the actual values of a new day's worth (48) of trade prices from an external source, obtains from the performance storage unit 23 the actual values of each trade price for the past M years (for example, the past 5 years), including the actual value of the new day's trade price, and each explanatory variable used to calculate the predicted trade price corresponding to each actual trade price for the past M years. Based on the obtained values, the update means 24 updates the algorithm of the calculation means 11 for calculating the predicted trade price. In this embodiment, the algorithm is updated once a day, but it is not limited to this.
[0042] Furthermore, the calculation means 11 has three algorithms, each used to calculate predicted transaction prices for time slots belonging to different time periods. For example, the first algorithm calculates predicted transaction prices for a total of 14 time slots: 8 slots from 5:30 to 9:00 and 6 slots from 17:00 to 19:30. The second algorithm calculates predicted transaction prices for 15 time slots from 9:30 to 16:30. The third algorithm calculates predicted transaction prices for 19 time slots from 20:00 to 5:00 the following day.
[0043] The update means 24 updates the first, second, and third algorithms of the calculation means 11 using values for each corresponding frame (values such as temperature-dependent power usage prediction and solar power generation prediction). Since it has been confirmed that the trend of execution prices differs depending on the time of day, by providing multiple algorithms as in this embodiment, the calculation means 11 can stably calculate accurate predicted values of execution prices.
[0044] Furthermore, as shown in Figure 5, the minimum price expected value derivation means 22 obtains from the performance storage unit 23 the values corresponding to the "predicted amount of solar power generation for the kth frame of the prediction target day" and the "coefficient obtained by comparing the predicted amount of solar power generation for the kth frame of the prediction target day with the average value of the predicted amount of solar power generation for the kth frame of the same month in the past" in Equation 3 when determining the minimum price expected value using Equation 3. These values are the predicted amount of solar power generation for each frame of the prediction target day, which was determined by the solar power generation prediction means 19 on the day before the prediction target day, and the predicted amount of solar power generation for each frame, which was determined by the solar power generation prediction means 19 on each day of the same month in a predetermined past period (for example, if the previous day is March 2, 2022, then each day of March for the five years from 2017 to 2021 and March 1, 2022).
[0045] In this embodiment, both temperature-dependent power consumption forecast and solar power generation forecast are used as explanatory variables. However, it is also possible to use either temperature-dependent power consumption forecast or solar power generation forecast as an explanatory variable and leave the other one blank. For example, temperature-dependent power consumption forecast may be used as an explanatory variable, and the average value of the predicted solar radiation at each point in the target area may be used instead of the predicted solar power generation forecast. Alternatively, solar power generation forecast may be used as an explanatory variable, and the average value of the predicted temperature at each point in the target area may be used instead of the temperature-dependent power consumption forecast.
[0046] Furthermore, if temperature-dependent power consumption forecasts are not used as explanatory variables, the derivation of temperature-dependent power consumption forecasts is unnecessary, and the power consumption forecasting means 15, etc., used in the derivation of temperature-dependent power consumption forecasts can be omitted. If solar power generation forecasts are not used as explanatory variables, the derivation of solar power generation forecasts is unnecessary, and the solar power generation forecasting means 19, etc., used in the derivation of solar power generation forecasts can be omitted.
[0047] However, we have confirmed that when both temperature-dependent electricity usage forecasts and solar power generation forecasts are used as explanatory variables, the accuracy of the transaction price forecast is higher compared to when only one of them is used as an explanatory variable and the other is not. Furthermore, in this embodiment, it is possible to choose whether or not to include the power generation capacity of the power generation equipment scheduled to be shut down at the power plant on the target date, calendar information, and the expected minimum price as explanatory variables.
[0048] Furthermore, the power contract price prediction method according to one embodiment of the present invention, which is applied to the power contract price prediction device 10, is a method for predicting the contract price of buying and selling electricity in a target area determined in the wholesale power market for each of several time slots into which a day is divided into predetermined unit time intervals, and uses either or both of the following as explanatory variables: a temperature-dependent predicted amount of power usage derived by weighting either the population of each district in the target area and the predicted temperature of each district in the target area with the other, and a predicted amount of solar power generation obtained by weighting either the amount of solar power generation installed in each district of the target area and the predicted amount of solar radiation of each district in the target area obtained from the status of solar power generation installation, with the predicted value of the contract price, which is the objective variable, to be determined by machine learning. [Examples]
[0049] Next, we will describe the experiments conducted to confirm the effects of the present invention. In the first experiment, the correlation coefficient between the temperature-dependent predicted electricity usage amount obtained by the electricity usage prediction method and the actual value of the contract price was compared with the correlation coefficient between the predicted temperature in Fukuoka City and the actual value of the contract price in the Kyushu area.
[0050] The correlation coefficient was calculated between the highest values (31 values for 31 days) of temperature-dependent predicted electricity usage in the Kyushu area during the morning and evening hours (5:30 to 9:00 and 17:30 to 19:30 each day in August 2019) and the highest values (31 values for 31 days) of the actual contract price in the Kyushu area during the morning and evening hours each day in the same month of August 2019. The correlation coefficient was -0.55. For reference, Figure 6(A) shows the 31 coordinates, with the highest values of the temperature-dependent predicted electricity usage as the X-coordinate and the highest values of the actual contract price as the Y-coordinate.
[0051] In contrast, when we calculated the correlation coefficient between the predicted maximum temperature in Fukuoka City during the morning and evening hours each day in August 2019 (31 values for 31 days) and the highest actual transaction price in the Kyushu area during the morning and evening hours each day in the same month of the same year (31 values for 31 days), the correlation coefficient was -0.45. For reference, Figure 6(B) shows a total of 31 coordinates, with the predicted maximum temperature in Fukuoka City as the X-coordinate and the highest actual transaction price as the Y-coordinate.
[0052] The experimental results showed that the highest predicted temperature-dependent electricity usage in the Kyushu area during the morning and evening hours, and the highest actual transaction price in the Kyushu area during the morning and evening hours, were more strongly correlated than the predicted maximum temperature in Fukuoka City during the morning and evening hours, and the highest actual transaction price in the Kyushu area during the morning and evening hours. Furthermore, the temperature has a significant impact on the trading price during the morning and evening hours of summer, which is thought to be due to the use of air conditioning after waking up or returning home.
[0053] In the second experiment, the correlation coefficient between the predicted solar power generation amount calculated by the solar power generation prediction method and the actual value of the contract price was compared with the correlation coefficient between the predicted solar radiation amount in Fukuoka City and the actual value of the contract price in the Kyushu area.
[0054] The correlation coefficient was calculated between the highest predicted solar power generation amount in the Kyushu area during the daytime hours of 9:30 to 17:00 each day in May 2019 (hereinafter simply referred to as "daytime hours") (31 values for 31 days) and the lowest actual transaction price in the Kyushu area during the daytime hours of each day in the same month of the same year (31 values for 31 days). The correlation coefficient was -0.69. For reference, Figure 7(A) shows the 31 coordinates, with the highest predicted solar power generation amount as the X coordinate and the lowest actual transaction price as the Y coordinate.
[0055] In contrast, when we calculated the correlation coefficient between the highest values of the predicted solar radiation for Fukuoka City during the daytime hours on each day in May 2019 (31 values for 31 days) and the lowest values of the actual transaction prices in the Kyushu area during the daytime hours on each day in the same month of the same year (31 values for 31 days), the correlation coefficient was -0.26. For reference, Figure 7(B) shows a total of 31 coordinates, with the highest value of the predicted solar radiation for Fukuoka City as the X coordinate and the lowest value of the actual transaction price as the Y coordinate.
[0056] The experimental results showed that the highest predicted solar power generation amount in the Kyushu area during daytime hours and the highest actual transaction price in the Kyushu area during daytime hours were more strongly correlated than the highest predicted solar radiation amount in Fukuoka City during daytime hours and the lowest actual transaction price in the Kyushu area during daytime hours. Furthermore, during the daytime hours in May, the transaction price tends to decrease due to the increase in solar power generation.
[0057] In the third experiment, we compared the predicted execution price with the actual execution price. Figure 8 shows the experimental results of predicting the transaction price in the Kyushu area on March 23, 2021, using an example where a random forest algorithm was employed as the calculation method. In this example, the explanatory variables for the target date of March 23, 2021, were the temperature-dependent predicted power usage, predicted solar power generation, the power generation capacity of power generation equipment scheduled to be shut down at power plants, calendar information, and the expected minimum price.
[0058] In Figure 8, the predicted value is the predicted execution price, and the actual value is the execution price for the Kyushu area determined by the Japan Electric Power Exchange on March 23, 2021. The average error between the predicted and actual values was 0.8 yen.
[0059] Although embodiments of the present invention have been described above, the present invention is not limited to the above-described forms, and any changes to the conditions, etc., that do not depart from the gist of the invention are all within the scope of application of the present invention. For example, the calculation algorithm does not necessarily need to have three components; it could be one, two, or four or more. However, it has been confirmed that when only one calculation algorithm is used, the accuracy of predicting the execution price is lower compared to when multiple algorithms corresponding to different time periods are used.
[0060] Furthermore, the means for obtaining predicted temperatures may be obtained from an external source for predicted temperatures for each district (e.g., each city, town, or village) within the target area, rather than for predicted temperatures for each predetermined point within the target area. In this case, the means for converting predicted temperatures is unnecessary. Similarly, the means for obtaining predicted solar radiation may be obtained from an external source for predicted solar radiation for each district (e.g., each city, town, or village) within the target area, rather than for predicted solar radiation for each predetermined point within the target area. In this case, the means for converting predicted solar radiation is omitted.
[0061] When the predicted temperature acquisition means obtains predicted temperatures for each point within the target area, the predicted temperature conversion means can derive the predicted temperatures for each district within the target area from the predicted temperatures for each point within the target area using a system different from the geographic information system. A different system might be, for example, a system that uses the predicted temperature of the point closest to the center of the district as the predicted temperature for that district. Similarly, when the solar radiation prediction acquisition means obtains solar radiation predictions for each point within the target area, the solar radiation prediction conversion means can also derive the solar radiation predictions for each district within the target area from the solar radiation predictions for each point within the target area using a system different from the information system.
[0062] Furthermore, the frequency of obtaining predicted temperatures within the target area from external sources using the predicted temperature acquisition means, the frequency of obtaining predicted solar radiation within the target area from external sources using the predicted solar radiation acquisition means, and the frequency of updating the algorithm of the calculation means using the update means can be adjusted according to the circumstances. [Explanation of Symbols]
[0063] 10: Electricity contract price prediction device, 11: Calculation means, 12: Predicted temperature acquisition means, 13: Population memory unit, 14: Predicted temperature conversion means, 15: Electricity usage prediction means, 16: Solar radiation prediction amount acquisition means, 17: Solar power generation introduction amount storage unit, 18: Solar radiation prediction amount conversion means, 19: Solar power generation amount prediction means, 20: Stopped power generation capacity derivation means, 21: Calendar storage unit, 22: Minimum price expected value derivation means, 23: Actual performance storage unit, 24: Update means
Claims
1. In a power contract price prediction device that predicts the contract price of electricity buying and selling in a target area, which is determined in the wholesale electricity market, for each of several time slots into which a day is divided into predetermined unit time intervals, The system includes a calculation means for calculating a predicted value of the transaction price, which is the objective variable, by machine learning, using either or both of the following as explanatory variables: a temperature-dependent predicted amount of electricity use derived by weighting either the population of each district in the target area and the predicted temperature of each district in the target area with the other, and a predicted amount of solar power generation obtained by weighting either the amount of solar power generation installed in each district of the target area and the predicted amount of solar radiation in each district of the target area, which are obtained from the status of installation of solar power generation equipment. The power contract price prediction device is characterized in that the machine learning uses an algorithm that calculates an updated predicted value of the contract price based on each actual value of the past contract price and each explanatory variable used to calculate the predicted value of the contract price corresponding to each actual value.
2. The electricity contract price prediction device according to claim 1 further comprises: a prediction temperature acquisition means for obtaining prediction temperatures for each of a plurality of predetermined locations within the target area from an external source; and a prediction temperature conversion means for deriving prediction temperatures for each district of the target area from the prediction temperatures of each point within the target area using a geographic information system, wherein the calculation means uses the temperature-dependent predicted amount of electricity usage as the explanatory variable.
3. The electricity contract price prediction device according to claim 1 further comprises: a solar radiation prediction amount acquisition means for obtaining the solar radiation prediction amount for each of a plurality of predetermined points within the target area from an external source; and a solar radiation prediction amount conversion means for deriving the solar radiation prediction amount for each district of the target area from the solar radiation prediction amount of each point within the target area using a geographic information system, wherein the calculation means uses the solar power generation prediction amount as the explanatory variable.
4. The electricity contract price prediction device according to claim 3, characterized in that one of the explanatory variables is a variable that acts to cause the predicted value of the contract price to approach or be equal to the lowest price of the contract price set in the wholesale electricity market, depending on the magnitude of the derived predicted amount of solar power generation relative to the predicted amount of solar power generation for the same month and time frame derived in the past.
5. A power contract price prediction device according to any one of claims 1 to 4, further comprising an update means for updating the algorithm of the calculation means for calculating the predicted value of the contract price based on the explanatory variables used to calculate the predicted value of the contract price corresponding to the actual value of the contract price determined in the wholesale power market and the actual value of the contract price.
6. The electricity contract price prediction device according to claim 5, wherein the algorithm of the calculation means is a plurality of algorithms, each used to calculate a predicted value of the contract price for the time slots belonging to different time periods.
7. In a method for predicting electricity contract prices, which predicts the contract price for buying and selling electricity in a target area determined in the wholesale electricity market for each of several time slots into which a day is divided into predetermined unit time intervals, A software program on a computer The temperature-dependent predicted power consumption, derived by weighting either the population of each district in the target area or the predicted temperature of each district in the target area with the other, and the predicted solar power generation, obtained from the status of solar power generation installations in each district of the target area or the predicted solar radiation of each district in the target area with the other, are used as explanatory variables, and the predicted value of the transaction price, which is the dependent variable, is obtained by machine learning. The method for predicting electricity contract prices is characterized by using an algorithm that calculates an updated predicted value of the contract price based on each actual value of the past contract price and each explanatory variable used to calculate the predicted value of the contract price corresponding to each actual value.