Electricity market price prediction system

Abstract

We provide an electricity market price forecasting system that can accurately predict electricity market prices. [Solution] The electricity market price forecasting system 1 is an electricity market price forecasting system that forecasts the electricity market price for a predetermined period (from 0:00 to 24:00), and comprises an acquisition unit 11 that uses a forecasting model to acquire predicted values ​​of the electricity market price every 30 minutes (a predetermined time period) during the predetermined period (from 0:00 to 24:00), and a calculation unit 12 (correction unit) that corrects the predicted values ​​using event information relating to events that affect the electricity market price.

Description

Technical Field

[0001] The present invention relates to a technology of a power market price prediction system for predicting power market prices.

Background Art

[0002] Conventionally, the technology of a power market price prediction system for predicting power market prices has been known. For example, it is as described in Patent Document 1.

[0003] Patent Document 1 discloses a power market price prediction device that predicts a power market price based on an estimated power supply amount by renewable energy on a prediction target day, actual value information of the power supply amount, and actual value information of the power market price. Thus, in the technology described in Patent Document 1, by predicting the estimated power supply amount and power demand amount that have a large influence on the power market price, the prediction accuracy of the power market price is improved. By improving the prediction accuracy of the power market price, the improvement of the electricity sales revenue can be achieved.

[0004] Here, when a sudden event such as a change in natural conditions that is difficult to predict occurs, a sudden price fluctuation of the power market price may occur due to the event. In the technology described in Patent Document 1, in order to reduce the prediction error of the power market price, the part of the sudden price fluctuation is excluded from the actual value information of the power market price.

[0005] However, since sudden price fluctuations may also occur in the future, simply excluding the part of the sudden price fluctuation from the actual value information of the power market price may conversely reduce the prediction accuracy of the power market price. Therefore, a technology for predicting the power market price more accurately is desired.

Prior Art Documents

Patent Documents

[0006]

Patent Document 1

Summary of the Invention

[0007] This invention was made in view of the above circumstances, and the problem it aims to solve is to provide an electricity market price forecasting system that can accurately predict electricity market prices. [Means for solving the problem]

[0008] The problems that this invention aims to solve are as described above, and the means for solving these problems will now be explained.

[0009] In other words, claim 1 provides an electricity market price forecasting system for forecasting electricity market prices for a predetermined period, comprising: an acquisition unit that uses a forecasting model to acquire predicted values ​​of the electricity market prices for each predetermined time period within the predetermined period; and a correction unit that uses event information relating to events affecting the electricity market prices to correct the predicted values.

[0010] In claim 2, the event information includes the price of power generation fuel at a predetermined past timing, and the correction unit performs a first correction based on the price of power generation fuel and the historical value of the electricity market price during the pre-daytime period within the predetermined period.

[0011] In claim 3, the correction unit performs the first correction during all time periods of the predetermined period.

[0012] In claim 4, the event information includes a forecast of solar power generation and a forecast of electricity demand in the area covered by the electricity market price, and the correction unit performs a second correction on the first corrected forecast value based on the solar power generation forecast, the electricity demand forecast, and past performance values ​​of the electricity market price during the daytime hours of the predetermined period.

[0013] Claim 5 includes a first prediction model created based on a first explanatory variable that includes historical values ​​of the electricity market price, and a second prediction model created based on a second explanatory variable that does not include historical values ​​of the electricity market price, wherein the acquisition unit uses the first prediction value obtained by the first prediction model as the prediction value of the electricity market price, and, when certain conditions are met, uses the second prediction value obtained by the second prediction model instead of the first prediction value during daytime hours within the predetermined period.

[0014] In claim 6, the predetermined conditions are set based on a solar power generation forecast and a power demand forecast.

[0015] In claim 7, the event information includes a solar power generation forecast and a power demand forecast, and the correction unit performs a third correction on the second forecast value used as the forecast value, based on the solar power generation forecast, the power demand forecast, and past performance values ​​of the power market price. [Effects of the Invention]

[0016] The present invention provides the following effects:

[0017] In this invention, electricity market prices can be predicted with high accuracy. [Brief explanation of the drawing]

[0018] [Figure 1] A block diagram showing the configuration of an electricity market price prediction system according to one embodiment of the present invention. [Figure 2] This diagram shows the overall structure of the market price prediction process according to this embodiment. [Figure 3] This diagram illustrates how the predicted values ​​from two different prediction models are combined. [Figure 4] A flowchart illustrating the market price prediction process. [Figure 5] Diagram illustrating the first correction. [Figure 6] Explanatory drawing of the second amendment. [Figure 7] Drawing showing a specific example of the second amendment. [Figure 8] Explanatory drawing of the third amendment. [Figure 9] Drawing showing a specific example of the third amendment.

MODE FOR CARRYING OUT THE INVENTION

[0019] Hereinafter, a power market price prediction system 1 according to an embodiment of the present invention will be described.

[0020] A power market price prediction system 1 according to an embodiment of the present invention predicts the power market price (hereinafter simply referred to as "power market price") of a target area (region) determined in the wholesale power market. The power market price prediction system 1 is configured by, for example, a general computer equipped with a CPU (Central Processing Unit). The power market price prediction system 1 includes an acquisition unit 11 and a calculation unit 12 (see FIG. 1). The acquisition unit 11 and the calculation unit 12 are realized as functions of hardware or software that constitutes the power market price prediction system 1.

[0021] Here, the main wholesale power markets are the day-ahead market (spot market) and the same-day market (hour-ahead market). The day-ahead market is a market where electricity transactions are conducted on the day before the actual supply and demand. In the day-ahead market, transactions are conducted for 48 products (48 frames) obtained by dividing one day into 30-minute units. On the other hand, the same-day market is a market where transactions are conducted until one hour before the actual supply and demand, and mainly takes charge of adjusting the excess or deficiency of the day-ahead market. In the present embodiment, the power market price prediction system 1 predicts the wholesale power market price in the day-ahead market (spot market).

[0022] Hereinafter, the day on which bids are made in the wholesale power market is referred to as "the current day", and the day of actual supply and demand (that is, the day following the day on which bids are made in the day-ahead market) is referred to as "the target day". Here, in the wholesale power market, for example, it is necessary to submit bids by 10:00 on the current day.

[0023] In this embodiment, the predicted values ​​obtained by the market price forecasting process (see Figure 4), described later, are used for planning bids. That is, once the predicted value of the electricity market price is obtained by the market price forecasting process, the user can predict the bid curve (supply and demand curve) for bidding in the wholesale electricity market. The market price forecasting process is executed in the morning of the day the bids are to be made (for example, at 7:00).

[0024] The configuration of the electricity market price forecasting system 1 will be described in detail below. As shown in Figure 1, the electricity market price forecasting system 1 comprises an acquisition unit 11, a calculation unit 12, and a display unit 13.

[0025] The acquisition unit 11 acquires various types of information. The acquisition unit 11 can acquire various types of information necessary for predicting electricity market prices from a predetermined storage unit of the computer or from an external source via a network such as the Internet.

[0026] The calculation unit 12 performs various calculations. The calculation unit 12 can execute various processes necessary for predicting electricity market prices. These various processes include the market price prediction process described later.

[0027] The display unit 13 is used to display various information. The display unit 13 is composed of a general-purpose display or the like. The display unit 13 can display various information in accordance with the instructions of the calculation unit 12.

[0028] Here, electricity market prices can experience rapid price fluctuations due to unforeseen events such as changes in natural conditions that are difficult to predict. To reduce the prediction error of electricity market prices in such cases, methods are known to exclude the portion of rapid price fluctuations from actual electricity market price data. However, since rapid price fluctuations may occur again in the future, simply excluding the portion of rapid price fluctuations from actual electricity market price data may actually decrease the accuracy of electricity market price predictions.

[0029] Therefore, the electricity market price forecasting system 1 according to this embodiment aims to improve the accuracy of electricity market price forecasts by performing market price forecasting processing, while taking into account the possibility of rapid price fluctuations occurring in the future.

[0030] First, we will explain the overview of the market price prediction process performed by the calculation unit 12 using Figures 2 and 3.

[0031] In the market price prediction process, two types of prediction models with different characteristics are generated by machine learning. Of the two prediction models, one (base model M1) predicts the base portion of the final predicted value F (see Figure 3) obtained by the market price prediction process. The other (daytime-only model M2) is specialized for daytime hours. The predicted values ​​of base model M1 and daytime-only model M2 are corrected according to their respective characteristics (for example, considering rapid price fluctuations). In this embodiment, the predicted value of base model M1 is corrected twice (first correction and second correction). The predicted value of daytime-only model M2 is corrected once (third correction).

[0032] After applying corrections to the predictions of both the base model M1 and the daytime-only model M2, the two sets of predictions are then combined into one. Specifically, the daytime portion of the base model M1's prediction is replaced with the prediction from the daytime-only model M2. Once a single prediction is created from these two sets of predictions, this single prediction is used as the final prediction F. In this way, when the market price prediction process is executed, multiple prediction models corresponding to different time periods of the day are used, and the results of machine learning can be corrected, resulting in highly accurate predictions.

[0033] The following sections will describe in detail the market price prediction process performed by the calculation unit 12, using Figures 2 to 9. Figure 4 is a flowchart illustrating the market price prediction process.

[0034] In step S11, the calculation unit 12 acquires weather information via the acquisition unit 11 and performs power generation forecasting and power demand forecasting. Here, "power generation forecasting" refers to, for example, the forecast of the amount of power generated by solar power generation facilities in the target area on a target day. "Power demand forecasting" refers to, for example, the forecast of the amount of power demand in the target area on a target day. The calculation unit 12 generates a forecasting model using machine learning in order to perform power generation forecasting and power demand forecasting.

[0035] Specifically, the calculation unit 12 acquires weather information for the target day distributed by a designated weather service provider. This weather information includes, for example, temperature, solar radiation, and wind speed for the target day. The calculation unit 12 performs machine learning to generate a prediction model with weather information as the explanatory variable and power generation and electricity demand as the dependent variable. The calculation unit 12 then inputs the acquired weather information into the prediction model and calculates predicted values ​​for power generation and electricity demand. The calculation unit 12 can also acquire predicted values ​​for power generation and electricity demand from the designated weather service provider without performing machine learning (i.e., without performing power generation and electricity demand predictions).

[0036] After performing the processing in step S11, the arithmetic unit 12 proceeds to step S12.

[0037] In step S12, the calculation unit 12 makes a first prediction of the electricity market price. Here, the "first prediction" is a prediction of the base portion of the final predicted value F (see Figure 3) obtained by the market price prediction process. That is, the calculation unit 12 calculates the final predicted value F of the electricity market price by making predetermined corrections and changes to the predicted value of the first prediction, as will be described later. For the first prediction of the electricity market price, the calculation unit 12 generates a prediction model (hereinafter referred to as "base model M1") using machine learning.

[0038] Specifically, the calculation unit 12 learns the necessary information using machine learning and generates a base model M1 with the electricity market price as the target variable. The calculation unit 12 learns from data from the most recent three months. For machine learning, deep learning is used, for example, to enable the prediction of a relatively broad price range.

[0039] The explanatory variables of the base model M1 include (a) weather information (temperature, solar radiation, and wind speed), (b) the previous day's electricity market price, (c) the result of subtracting the predicted power generation amount from the predicted power demand amount, (d) the average price by time of day in the previous day's electricity market price, (e) the average price of the electricity market price over the past few days, (f) the fuel price from the past few days, (g) the planned shutdown amount of the power plant, and (h) calendar information including the date and day of the week. The explanatory variables of the base model M1 are acquired every 48 frames. The above explanatory variables are examples of the first explanatory variables according to the present invention.

[0040] Thus, the calculation unit 12 generates a base model M1 and calculates the electricity market price (the predicted value of the first forecast) by performing calculations using the base model M1. Hereinafter, the predicted value of the first forecast will be referred to as the "first predicted value." The first predicted value is acquired at predetermined time intervals, and in this embodiment, it is acquired every 48 30-minute intervals from the end of the day.

[0041] After performing the processing in step S12, the arithmetic unit 12 proceeds to step S13.

[0042] In step S13, the calculation unit 12 performs a first correction on the first predicted value of the electricity market price. Here, the "first correction" is a correction performed based on past actual values ​​of the electricity market price during the pre-daytime hours and fuel prices from a predetermined period ago. In this embodiment, the pre-daytime hours refer to the time of day when electricity demand is lowest, such as before sunrise. In other words, the inventors of this application have found that in the time of day when there is little power generation and electricity demand is lowest, the trend of fuel prices from a predetermined period ago influences the electricity market price. The reason for focusing on fuel prices from a predetermined period ago is to take into account the time lag between the purchase of fuel and its actual use. Therefore, in this embodiment, the first predicted value is corrected based on the correlation between the electricity market price from 0:00 to 6:00 (the time of day when there is little solar power generation and electricity demand is lowest) and the fuel prices from the most recent few days ago (a shift of several days in fuel prices).

[0043] It is known that the electricity market price between 0:00 and 6:00 affects the electricity market price in subsequent hours. That is, if the electricity market price between 0:00 and 6:00 is high, the electricity market price in subsequent hours is likely to be high as well. Conversely, if the electricity market price between 0:00 and 6:00 is low, the electricity market price in subsequent hours is likely to be low as well. Therefore, in this embodiment, in the first correction, the calculation unit 12 performs a correction not only for 0:00 to 6:00 but also for subsequent hours (6:00 to 24:00). In other words, the first correction targets all 48 timeframes.

[0044] Figure 5 is an explanatory diagram of the first correction. The graph in Figure 5 shows the relationship between the average historical value of the electricity market price from 0:00 to 6:00 and the fuel price from the most recent few days prior. In other words, each plot in Figure 5 shows the average historical value of the electricity market price from 0:00 to 6:00 relative to the fuel price from the most recent few days prior. The straight line in Figure 5 shows the approximate equation (y=0.80x+1.68) that shows the relationship between the fuel price from the most recent few days prior, calculated from the above plots, and the average historical value of the electricity market price from 0:00 to 6:00.

[0045] Thus, when performing the first correction, as shown in the [Correction Flow] in Figure 5, the calculation unit 12 first refers to the fuel price from the most recent few days ago and substitutes the obtained value into x in the above approximation formula to calculate y (hereinafter referred to as the "first correction price"). Next, the calculation unit 12 substitutes the calculated first correction price into each of the 48 time slots from 0:00 to 5:30. In this way, the same value (the first correction price) is substituted into each time slot from 0:00 to 5:30.

[0046] The calculation unit 12 then calculates the average difference between the first correction price substituted into each time slot from 0:00 to 5:30 and the first predicted value for each time slot. The calculation unit 12 then adds the calculated average difference to the first predicted value for each time slot from 6:00 to 23:30 (the time period after 0:00 to 6:00) out of the 48 time slots. If the calculated average difference is negative, the addition to the first predicted value for each time slot effectively becomes a subtraction (the predicted electricity market price becomes lower). This completes the first correction.

[0047] In the following section, referring to the [Example] in Figure 5, we will explain a specific example of calculating the first adjustment price with respect to the first adjustment.

[0048] When calculating the first adjustment price, the calculation unit 12 first obtains the fuel price from the most recent few days ago, as described above. In this embodiment, the calculation unit 12 calculates the fuel price from the most recent few days ago (LNG price in this embodiment) using the following equation 1.

[0049] (Math 1) Fuel price from the most recent few days ago = (Exchange rate × JKM) / 160

[0050] JKM is the spot price index for LNG. 160 is a predetermined constant. For example, if the exchange rate is 149.94 yen and JKM is 18.745, calculating 149.94 yen × 18.745 / 160 gives 17.57 yen as the fuel price from a few days ago. Substituting this calculated 17.57 yen into the above approximation formula gives 15.73 yen as the first adjustment price to be substituted into each time slot from 0:00 to 5:30.

[0051] Thus, the calculation unit 12 calculates the first correction price and then uses the calculated first correction price to perform the first correction on the first predicted value.

[0052] After performing the processing in step S13, the arithmetic unit 12 proceeds to step S14.

[0053] In step S14, the calculation unit 12 performs a second correction on the first predicted value after the first correction (hereinafter referred to as the "first corrected predicted value"). Here, the "second correction" is a correction performed based on the relationship between power generation and electricity demand. More specifically, the "second correction" is a correction performed on the first corrected predicted value during daytime hours (15:00 to 20:00, as described later), based on the power generation forecast, electricity demand forecast, and past performance values ​​of the electricity market price. In other words, the inventors of the present invention have found that in the time period when power generation is expected to gradually decrease and electricity demand is generally expected to gradually increase, the relationship between power generation and electricity demand affects the electricity market price.

[0054] For example, during a period when power generation is gradually decreasing and electricity demand is gradually increasing, if power generation is extremely low relative to electricity demand, electricity market prices may become unstable and could surge sharply. Therefore, the calculation unit 12 considers the possibility of a sharp surge in electricity market prices based on the relationship between power generation and electricity demand during the aforementioned period from 3 PM to 8 PM, and performs a second correction on the first corrected forecast value.

[0055] In this embodiment, the relationship between power generation and electricity demand is measured by the difference between the average difference in predicted values ​​between electricity demand and power generation across all 48 time slots from 0:00 to 23:00, and the difference in predicted values ​​between electricity demand and power generation in each time slot (hereinafter referred to as "difference from the average D") (see graph in Figure 6). In the following, "average difference in predicted values ​​between electricity demand and power generation across all 48 time slots" may be referred to as "average difference in predicted values".

[0056] Furthermore, the inventors of this application have found that there is a correlation between the average of the predicted electricity market prices from 0:00 to 6:00, the difference D from the average, and the maximum electricity market price. Here, the maximum electricity market price is the maximum electricity market price out of all 48 timeframes. The reason for focusing on the period from 0:00 to 6:00 regarding the average electricity market price is that this time period is the start-up time for power plants and is also the time of day when electricity demand is the lowest, so it can be said to be the base period for the electricity market price on that day.

[0057] The table in Figure 6 shows an example of the relationship between the average predicted electricity market price from 0:00 to 6:00 during a peak demand period, the difference D from the average, and the maximum electricity market price (historical value). As shown in the table in Figure 6, for example, even if the average predicted value is about the same, if the difference D from the average is large, the maximum electricity market price (historical value) is likely to be high (see July 18th and 27th). Also, for example, even if the average predicted value is low, if the difference D from the average is large, the maximum electricity market price (historical value) is likely to be high (see July 12th and 18th). Furthermore, it can be seen that when the difference D from the average is 3,000 or more, a price increase of 20 yen or more occurs.

[0058] Thus, when the average electricity market price from 0:00 to 6:00 is high, and the difference D from the average is large, the maximum electricity market price for the target day tends to be high. In this case, the larger the difference D from the average, the greater the possibility of a sharp surge in the electricity market price for the target day. In other words, by making adjustments that take into account the relationship between power generation and electricity demand, it is possible to respond to fluctuations in high electricity market prices.

[0059] Therefore, in the second correction, the calculation unit 12 performs machine learning on the necessary information to generate a regression equation using multiple regression analysis, with the average of the predicted electricity market prices from 0:00 to 6:00 and the difference D from the average as explanatory variables, and the maximum value of the electricity market price as the dependent variable.

[0060] Thus, when performing the second correction, the calculation unit 12 first obtains the average of the predicted electricity market prices from 0:00 to 6:00 and the difference D from the average, based on the first corrected predicted values, and substitutes these into the regression equation to calculate the predicted value of the maximum value of the electricity market price (hereinafter referred to as the "second correction price"). The calculation unit 12 then calculates the difference between the calculated second correction price and the first corrected predicted value for each time slot, and adds the calculated value to the first corrected predicted value for each time slot.

[0061] Thus, in each time slot, the first corrected forecast value after the second correction (hereinafter referred to as the "second corrected forecast value") is adjusted to a higher price according to the difference D from the average. This completes the second correction. The second correction is applied to time slots between 15:00 and 19:30 where the difference between the predicted values ​​of electricity demand and generation exceeds the average difference between the predicted values.

[0062] Below, we will explain a specific example of the second correction using Figure 7.

[0063] Figure 7(a) overlays the predicted value (first adjusted predicted value) of the electricity market price for a specific day during the demand season with the actual value after the prediction, making it easy to compare them. As shown in Figure 7(a), there was a sharp rise in the electricity market price between 3 PM and 8 PM on this day, even though it was not predicted.

[0064] In contrast, Figure 7(b) shows the difference between the predicted electricity demand and generation for each time slot on the same target day, as well as the average of the differences in predicted values. As shown in Figure 7(b), during the 15:00-20:00 time slot on this target day, the difference between the predicted electricity demand and generation for each time slot exceeds the average of the differences in predicted values. Furthermore, as shown in Figure 7(a), this time slot corresponds to a period when there was a sharp rise in electricity market prices. Therefore, in this case, a second correction is applied to the time slot from 15:00 to 20:00.

[0065] Figure 7(c) shows the results of applying the second correction to the 15:00-19:30 time slot shown in Figure 7(b). Specifically, the difference between the calculated second correction price and the first corrected predicted value for each time slot from 15:00 to 19:30 is calculated, and this calculated value is added to the first corrected predicted value for each time slot. As a result, the second corrected predicted value is higher than the initial value and becomes almost the same as the actual value (see the dotted circle in Figure 7(c)). In this way, by applying the second correction, a predicted value that takes into account the possibility of a sudden surge in electricity market prices is obtained.

[0066] After performing the processing in step S14, the arithmetic unit 12 proceeds to step S15.

[0067] In step S15, the calculation unit 12 determines whether or not to make a second forecast of the electricity market price. Here, the "second forecast" is a forecast of the electricity market price that is specifically tailored to daytime hours. That is, the inventors of the present invention have found that during daytime hours, the relationship between power generation and electricity demand has a greater impact on the electricity market price than historical values ​​of the electricity market price. Therefore, in this embodiment, if certain conditions are met, a forecast value specifically tailored to the daytime hours is used during daytime hours, rather than a forecast value using the base model M1 as described above.

[0068] In this embodiment, whether or not a predetermined condition is met is determined based on the ratio of generated power to electricity demand. Specifically, in step S15, if the calculation unit 12 determines that the ratio of generated power to electricity demand exceeds a predetermined threshold, it proceeds to step S16. In this case, the calculation unit 12 uses a forecast value specific to daytime hours, as will be described later. On the other hand, in step S15, if the calculation unit 12 determines that the ratio of generated power to electricity demand does not exceed a predetermined threshold, it proceeds to step S19. In this case, the calculation unit 12 does not use a forecast value specific to daytime hours, as will be described later. That is, the second corrected forecast value is adopted as the final forecast value F.

[0069] Thus, the calculation unit 12 proceeds to step S16 when the ratio of generated power to electricity demand is relatively high and there is a high probability that the electricity market price will fluctuate to a low price, as will be described later. On the other hand, the calculation unit 12 proceeds to step S19 when the ratio of generated power to electricity demand is relatively low and there is a low probability that the electricity market price will fluctuate to a low price.

[0070] In step S16, the calculation unit 12 makes a second prediction of the electricity market price during daytime hours. Here, the "second prediction" is a prediction specifically for daytime hours, as described above. In this embodiment, the daytime hours are defined as the period from 6:00 to 15:00. In this embodiment, the calculation unit 12 makes predictions only for the period from 6:00 to 15:00 (daytime hours) and does not make predictions for other time periods. In this way, the calculation unit 12 reduces the processing load by making predictions only for the necessary time periods. For the second prediction of the electricity market price, the calculation unit 12 generates a machine learning prediction model (hereinafter referred to as the "daytime-only model M2").

[0071] Specifically, the calculation unit 12 learns the necessary information using machine learning and generates a daytime-only model M2 with the electricity market price as the target variable. The calculation unit 12 learns from data from the most recent three months. For machine learning, statistical methods such as multiple regression analysis are used.

[0072] The explanatory variables for the daytime-only model M2 include (a) weather information (temperature, solar radiation, and wind speed), (c) the result of subtracting the predicted power generation amount from the predicted power demand amount, (f) fuel prices from the most recent few days, (g) planned power plant shutdowns, and (h) calendar information.

[0073] Thus, the explanatory variables of the daytime-only model M2 differ from those of the base model M1 in that they do not include (b) the previous day's electricity market price, (d) the average price by time of day in the previous day's electricity market price, and (e) the average price of the electricity market price over the past few days. In other words, the explanatory variables of the daytime-only model M2 differ from those of the base model M1 in that they do not include historical data of the electricity market price. Note that the above explanatory variables are just one example of the second explanatory variables according to the present invention.

[0074] Thus, the calculation unit 12 generates a daytime-only model M2 and calculates the electricity market price (the predicted value of the second forecast) by performing calculations using the daytime-only model M2. Hereafter, the predicted value of the second forecast will be referred to as the "second forecast value." The second forecast value is obtained for each of the 48 time slots, corresponding to the time period from 6:00 to 15:00 (the 6:00 to 14:30 time slot).

[0075] After performing the processing in step S16, the arithmetic unit 12 proceeds to step S17.

[0076] In step S17, the calculation unit 12 performs a third correction on the second predicted value of the electricity market price. Here, the "third correction" is a correction that, like the second correction, is based on the relationship between the amount of power generated and the amount of electricity demand. In other words, the inventors of the present invention have found that during periods when the amount of power generated is expected to increase gradually and the amount of electricity demand is generally not expected to be very high (daytime hours), the relationship between the amount of power generated and the amount of electricity demand affects the electricity market price.

[0077] For example, during periods when power generation is expected to gradually increase and electricity demand is generally not expected to be very high, there is a possibility that power generation will be in surplus and electricity market prices will plummet sharply. Therefore, the calculation unit 12 considers the possibility of a sharp drop in electricity market prices based on the relationship between power generation and electricity demand during daytime hours (6:00 to 15:00) and applies a third correction to the second predicted value.

[0078] Here, Figure 8 shows an example of the relationship between the proportion of electricity generated to electricity demand and the electricity market price during periods of low demand, both on weekdays and weekends. In other words, each plot in Figure 8 shows the electricity market price in relation to the proportion of electricity generated to electricity demand. As shown in Figure 8, on weekdays, the electricity market price drops sharply when the proportion of electricity generated to electricity demand is around 40%. On weekends, the electricity market price drops sharply to 0.01 yen / kWh (a crash) when the proportion of electricity generated to electricity demand is around 30%. Considering these factors, in this embodiment, we focus on the proportion of electricity generated to electricity demand as the relationship between electricity generated and electricity demand. In other words, by making adjustments that take into account the relationship between electricity generated and electricity demand, it is possible to respond to fluctuations in the electricity market price to low levels.

[0079] Therefore, in the third correction, the calculation unit 12 performs machine learning on the necessary information to generate a regression equation using simple linear regression, with the average proportion of power generation to power demand in all time slots from 6:00 to 14:30 (hereinafter referred to as the "average proportion of power generation") and the proportion of power generation to power demand in each time slot (hereinafter referred to as the "proportion of power generation") as explanatory variables, and the electricity market price as the dependent variable.

[0080] Thus, when performing the third correction, the calculation unit 12 first obtains the average ratio of power generation from 6:00 to 15:00 and the ratio of power generation based on the second predicted values, and then calculates a predicted value for the electricity market price (hereinafter referred to as the "third correction price") by substituting these values ​​into the regression equation. The calculation unit 12 then calculates the difference between the calculated third correction price and the second predicted value for each time slot, and subtracts the calculated value from the second predicted value for each time slot.

[0081] Thus, in each time slot, the second forecast value after the third correction (hereinafter referred to as the "third corrected forecast value") is adjusted to a lower price according to the proportion of electricity generation that accounts for electricity demand. This completes the third correction. The third correction is applied to time slots between 6:00 and 14:30 where the proportion of electricity generation is below the average proportion of electricity generation.

[0082] Below, we will explain a specific example of the third correction using Figure 9.

[0083] Figure 9(a) overlays a predicted value (second prediction) for the electricity market price on a specific day during a period of low demand, with the actual value after the prediction, for easier comparison. As shown in Figure 9(a), during the 9:00-12:00 timeframe on this day, there was a sharp drop in the electricity market price, which was not predicted.

[0084] In contrast, Figure 9(b) shows the proportion of power generation for each time slot and the average proportion of power generation for the same target day. As shown in Figure 9(b), during the 9:00 to 15:00 time slot on this target day, the proportion of power generation for each time slot is below the average proportion of power generation. Furthermore, as shown in Figure 9(a), this time slot actually corresponds to the period when the electricity market price plummeted. Therefore, in this case, a third correction is applied to the time slots during the 9:00 to 15:00 time slot.

[0085] Figure 9(c) shows the results of applying a third correction to the 9:00-14:30 time slot shown in Figure 9(b). Specifically, the difference between the calculated third correction price and the second predicted value for each time slot from 9:00 to 14:30 is calculated, and this calculated value is subtracted from the second predicted value for each time slot. As a result, the predicted value after the third correction becomes lower than the initial value, and the predicted value after the third correction becomes almost the same as the actual value (see the dotted circle in Figure 9(c)). In this way, by applying the third correction, a predicted value that takes into account the possibility of a sudden collapse in electricity market prices is obtained.

[0086] After performing the processing in step S17, the arithmetic unit 12 proceeds to step S18.

[0087] In step S18, the calculation unit 12 combines the predicted value of the corrected base model M1 (the second corrected predicted value) and the predicted value of the corrected daytime-only model M2 (the third corrected predicted value). Specifically, the calculation unit 12 replaces the predicted value from 6:00 to 15:00 in the second corrected predicted value with the third corrected predicted value. In this way, the predicted values ​​from the two prediction models are combined to create a single predicted value. The calculation unit 12 uses this single predicted value as the final predicted value F.

[0088] After performing the processing in step S18, the arithmetic unit 12 proceeds to step S19.

[0089] In step S19, the calculation unit 12 displays the final predicted value F on the display unit 13. In this way, the user can confirm the final predicted value F through the display on the display unit 13.

[0090] After performing the processing in step S19, the calculation unit 12 terminates the market price prediction process.

[0091] As described above, the electricity market price prediction system 1 according to this embodiment is A power market price forecasting system that predicts the power market price for a predetermined period (from 0:00 to 24:00), An acquisition unit 11 that uses a prediction model to acquire predicted values ​​of the electricity market price every 30 minutes (a predetermined time period) during the predetermined period (from 0:00 to 24:00), A calculation unit 12 (correction unit) that corrects the predicted value using event information relating to events that affect the aforementioned electricity market price, It is equipped with the following features.

[0092] This configuration allows for accurate prediction of electricity market prices. Specifically, by applying corrections to the results of machine learning, it becomes possible to make predictions that take into account events that cannot be handled by machine learning alone, thus enabling more accurate predictions.

[0093] Furthermore, in the electricity market price forecasting system 1, The aforementioned event information is, Including the price of fuel for power generation from the most recent few days ago (a predetermined time in the past), The calculation unit 12 (correction unit) is: This involves making a first adjustment based on the aforementioned fuel price for power generation and the historical value of the aforementioned electricity market price during the pre-daytime hours (0:00 to 6:00) within the aforementioned predetermined period (0:00 to 24:00).

[0094] This configuration allows for accurate prediction of the final predicted value F, taking into account the characteristics of fuel prices for power generation.

[0095] Furthermore, in the electricity market price forecasting system 1, The calculation unit 12 (correction unit) is, The first correction (LNG correction) is performed during all time periods within the specified period (from 0:00 to 24:00).

[0096] This configuration allows for accurate prediction of the final forecast value F for the entire day, taking into account the characteristics of fuel prices for power generation.

[0097] Furthermore, in the electricity market price forecasting system 1, The aforementioned event information is, This includes solar power generation forecasts and electricity demand forecasts (see Figure 2, etc.) for the area covered by the aforementioned electricity market price, The calculation unit 12 (correction unit) is, A second correction is made to the first corrected forecast value, based on the solar power generation forecast, the electricity demand forecast, and past performance values ​​of the electricity market price, during the daytime hours of 15:00 to 20:00 within the predetermined period (0:00 to 24:00).

[0098] This configuration allows for accurate prediction of the final predicted value F, taking into account the characteristics of daytime hours.

[0099] Furthermore, in the electricity market price forecasting system 1, The aforementioned prediction model, The system includes a base model M1 (first forecast model) created based on a first explanatory variable that includes historical values ​​of the aforementioned electricity market price, and a daytime-only model M2 (second forecast model) created based on a second explanatory variable that does not include historical values ​​of the aforementioned electricity market price. The acquisition unit 11 is, As the predicted value of the aforementioned electricity market price, the first predicted value obtained by the base model M1 (first prediction model) is used, When certain conditions are met, the second predicted value obtained by the daytime-only model M2 (second predicted model) is used instead of the first predicted value during the daytime hours within the predetermined period (from 0:00 to 24:00).

[0100] This configuration utilizes multiple prediction models tailored to different time periods of the day, allowing for highly accurate prediction of the final predicted value F.

[0101] Furthermore, in the electricity market price forecasting system 1, The aforementioned predetermined conditions are: It is set based on solar power generation forecasts and electricity demand forecasts.

[0102] This reduces the processing load by creating a daytime-only model M2 (a second prediction model) as needed.

[0103] Furthermore, in the electricity market price forecasting system 1, The aforementioned event information is, Including solar power generation forecasts and electricity demand forecasts, The calculation unit 12 (correction unit) is, A third correction is made to the second predicted value used as the aforementioned predicted value, based on the solar power generation forecast, the electricity demand forecast, and past performance values ​​of the electricity market price.

[0104] This configuration allows for accurate prediction of the final predicted value F.

[0105] Furthermore, the information (data) used in the explanatory variables in each process according to this embodiment is one form of implementing the event information according to the present invention. Furthermore, the base model M1 according to this embodiment is an example of implementing the first prediction model according to the present invention. Furthermore, the daytime-only model M2 according to this embodiment is an example of implementing the second prediction model according to the present invention.

[0106] Although embodiments of the present invention have been described above, the present invention is not limited to the above configuration, and various modifications are possible within the scope of the invention as described in the claims.

[0107] For example, the processing performed by the arithmetic unit 12 does not necessarily have to be performed by the arithmetic unit 12 alone; it may be performed in cooperation with (outsourced to) the arithmetic unit of another system, for example. Also, the type of machine learning performed by the arithmetic unit 12 is not limited to that of this embodiment, and various types can be adopted.

[0108] Furthermore, the information used in the explanatory variables of each model in this embodiment is merely an example and can be changed as appropriate. Also, when using past performance values ​​for the most recent predetermined period as explanatory variables, the predetermined period can be changed as appropriate. In addition, the time periods to which the first, second, and third corrections apply can be changed as appropriate.

[0109] Furthermore, fuel prices refer to the prices of fuels used for power generation and are not limited to LNG prices. In other words, fuel prices may include the prices of various fuels such as crude oil and coal. [Explanation of Symbols]

[0110] 1. Electricity Market Price Prediction System 11 Acquisition Department 12 Arithmetic section

Claims

1. A power market price forecasting system that predicts the power market price for a predetermined period, An acquisition unit that uses a prediction model to acquire predicted values ​​of the electricity market price for each predetermined time period within the predetermined period, A correction unit that corrects the predicted value using event information relating to events that affect the electricity market price, Equipped with, Electricity market price prediction system.

2. The aforementioned event information is, Including past fuel prices for power generation at specified time points, The correction unit, A first adjustment is made based on the price of the fuel for power generation and the historical value of the electricity market price during the pre-daytime hours within the predetermined period. The electricity market price forecasting system according to claim 1.

3. The correction unit, The first correction is performed during all time periods within the predetermined period. The electricity market price forecasting system according to claim 2.

4. The aforementioned event information is, This includes forecasts of solar power generation and electricity demand in the area covered by the aforementioned electricity market price, The correction unit, A second correction is made to the first corrected forecast value based on the solar power generation forecast, the electricity demand forecast, and the historical electricity market price during the daytime hours of the predetermined period. The electricity market price forecasting system according to claim 3.

5. The aforementioned prediction model, The system includes a first forecasting model created based on a first explanatory variable that includes historical values ​​of the aforementioned electricity market price, and a second forecasting model created based on a second explanatory variable that does not include historical values ​​of the aforementioned electricity market price. The acquisition unit is, The first predicted value obtained by the first prediction model is used as the predicted value of the electricity market price, If certain conditions are met, during the daytime hours of the predetermined period, the second predicted value obtained by the second prediction model is used instead of the first predicted value. A power market price prediction system according to any one of claims 1 to 3.

6. The aforementioned predetermined conditions are: It is set based on solar power generation forecasts and electricity demand forecasts. The electricity market price forecasting system according to claim 5.

7. The aforementioned event information is, Including solar power generation forecasts and electricity demand forecasts, The correction unit, A third correction is made to the second predicted value used as the aforementioned predicted value, based on the solar power generation forecast, the electricity demand forecast, and the historical value of the electricity market price. The electricity market price forecasting system according to claim 5.

Images