Estimation program, estimation method and information processing device

Abstract

To provide an estimation program, an estimation method and an information processing device by which prediction accuracy of a contract price improves.SOLUTION: An estimation program causes a computer to execute processing of: generating a machine learning model by approximating each of a first curve and a second curve having one intersection with the first curve in a first period to a first approximation formula and a second approximation formula as a linear function passing through the intersection and performing learning by using inclinations and data of segments of the first approximation formula and the second approximation formula, respectively; and estimating an intersection between the first curve and the second curve in a second period equivalent to after the first period by using the learned machine learning model.SELECTED DRAWING: Figure 5

Description

[Technical field]

[0001] The present invention relates to an estimation program, an estimation method, and an information processing device. [Background technology]

[0002] Power exchanges act as intermediaries in the buying and selling of electricity between power generation companies and electricity retailers. Power exchanges mediate the buying and selling of electricity using a computer-based trading system. One of the transactions that take place at power exchanges is the spot market. In the spot market, the trading system divides a day into measurement units such as 30 minutes, and accepts selling bids for electricity from power generation companies and buying bids for electricity from retailers for each delivery time slot that is the length of the measurement unit. After the trading system finishes accepting orders, it determines the contract price based on the status of the selling bids and the buying bids.

[0003] The contract price and the contract quantity are determined based on a supply curve that shows the relationship between the quantity of sell orders and the price, and a demand curve that shows the relationship between the quantity of buy orders and the price. Therefore, if the supply curve and the demand curve for a specific delivery time period can be estimated in advance, the contract price can be accurately estimated.

[0004] Therefore, the following technologies are available as technologies related to electricity prices. For example, a technology has been proposed in which a demand curve and a supply curve are calculated using a sell contract rate function and a buy contract rate function obtained by performing spline regression on the actual values ​​of the contract price, with the sell contract rate and the buy contract amount as explanatory variables. A technology has also been proposed in which electricity prices are predicted using multiple forecasting methods, and the electricity price predicted by the forecasting method with the highest index value, which is an indicator of the prediction accuracy, is used as the predicted value. A technology has also been proposed in which electricity retail prices are predicted to provide a probabilistic evaluation of costs and risks. Another demand forecasting technology has been proposed in which a recurrent neural network model trained using multiple time-series demand observations generates a probabilistic demand forecast for a target item. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2016-33801 A [Patent Document 2] JP 2018-13934 A [Patent Document 3] US Patent Application Publication No. 2004 / 0215529 [Patent Document 4] U.S. Pat. No. 1,093,6947 Summary of the Invention [Problem to be solved by the invention]

[0006] However, in the conventional technology, the prediction accuracy of the demand curve or the supply curve for a future time period is low, and the prediction accuracy of the contract price determined based on the demand curve and the supply curve is insufficient. Note that the problem of insufficient prediction accuracy of the contract price also occurs in market transactions other than the electricity trading.

[0007] The disclosed technology has been made in consideration of the above, and aims to provide an estimation program, an estimation method, and an information processing device that improve the accuracy of predicting the contract price. [Means for solving the problem]

[0008] In one aspect of the estimation program, estimation method, and information processing device disclosed in the present application, a computer is caused to execute a process of approximating a first curve and a second curve having one intersection with the first curve during a first period to a first approximation formula and a second approximation formula, which are linear functions passing through the intersection, and learning the first approximation formula and the second approximation formula using data on the slope and intercept, respectively, to generate a machine learning model, and estimating an intersection between the first curve and the second curve during a second period that corresponds to after the first period using the trained machine learning model. Effect of the Invention

[0009] In one aspect, the present invention can improve the accuracy of predicting an execution price. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a block diagram of an agreement price estimation device according to a first embodiment. [Diagram 2] FIG. 2 is a diagram for explaining the generation of the initial function. [Diagram 3] FIG. 3 is a diagram showing an example of time-series changes in the slope and intercept values ​​of the demand initial function created by the initial function generating unit. [Figure 4] FIG. 4 is a diagram showing an example of time-series changes in the slope and intercept values ​​after smoothing. [Diagram 5] FIG. 5 is a diagram for explaining the creation of the demand approximation equation and the supply approximation equation. [Figure 6] FIG. 6 is a diagram illustrating an outline of the estimation process performed by the estimation unit. [Figure 7] FIG. 7 is a diagram illustrating a comparison of prediction errors between the estimation by the contract price estimation device according to the first embodiment and the estimation using the LSTM alone. [Figure 8] FIG. 8 is a flowchart of the learning process of the demand curve model and the supply curve model in the learning phase. [Figure 9] FIG. 9 is a flowchart of the estimation process by the contract price estimation device in the estimation phase. [Figure 10] FIG. 10 is a hardware configuration diagram of the contract price estimation device according to the first embodiment. [Figure 11] FIG. 11 is a diagram illustrating an example of a configuration of a computer system according to the second embodiment. [Figure 12] FIG. 12 is a block diagram of a bidding computer. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, an embodiment of the estimation program, the estimation method, and the information processing device disclosed in the present application will be described in detail with reference to the drawings. Note that the estimation program, the estimation method, and the information processing device disclosed in the present application are not limited to the following embodiment. EXAMPLES

[0012] Fig. 1 is a block diagram of an agreement price estimation device according to a first embodiment. As shown in Fig. 1, the agreement price estimation device 1 according to the present embodiment includes a data storage unit 11, an approximation unit 12, a model generation unit 13, a model storage unit 14, and an estimation unit 15. The agreement price estimation device 1 has two operation phases: a learning phase in which the demand curve model 142 and the supply curve model 143 in the model storage unit 14 perform learning, and a learning phase in which the agreement price is learned using the learned demand curve model 142 and the supply curve model 143.

[0013] The data storage unit 11 stores information on the contract price, demand curve, and supply curve for each of the past 30-minute time slots in the spot market, for example. The data storage unit 11 may also store, for example, the total contract amount, the selling bid price, and the buying bid price. In addition, the data storage unit 11 may also store, for example, information on the temperature, the day of the week, and the power usage rate for each time slot. Hereinafter, each 30-minute time slot will be referred to as a measurement unit time slot.

[0014] Here, for a first time period, which is a specific measurement time period, and a second time period, which is a measurement unit time period preceding the first time period, a data set including the demand curve and supply curve for the first time period and the contract price for the second time period is used to train the demand curve model 142 and the supply curve model 143.

[0015] The model storage unit 14 has an agreement price model 141, a demand curve model 142, and a supply curve model 143, which are machine learning models.

[0016] The demand curve model 142 is a machine learning model that inputs the contract price in the second time zone and outputs an estimate of a value that defines an approximation equation of the demand curve in the first time zone. In this embodiment, the demand curve model 142 has a demand slope model 142a and a demand intercept model 142b.

[0017] The demand slope model 142a is a machine learning model that inputs the contract price in the second time slot, which is time series data, and outputs an estimate of the slope of a demand approximation equation that approximates the demand curve in the first time slot for each second time slot to a linear function. This demand approximation equation is an example of a "first approximation equation."

[0018] For example, the demand gradient model 142a is a model using a recurrent neural network (RNN). More specifically, the demand gradient model 142a can use a long short-term memory (LSTM) network, which is a type of RNN. LSTM excels at learning and prediction (regression and classification) of time-series data.

[0019] The demand intercept model 142b is a machine learning model that inputs the contract price in the second time slot, which is time series data, and outputs an estimated value of the intercept of a demand approximation equation that approximates the demand curve in the first time slot for each second time slot to a linear function. For example, the demand intercept model 142b is a model that uses an RNN. More specifically, the demand intercept model 142b can use an LSTM network.

[0020] The supply curve model 143 is a machine learning model that inputs the contract price in the second time zone and outputs an estimate of a value that defines an approximation equation of the supply curve in the first time zone. This supply approximation equation is an example of a "second approximation equation." In this embodiment, the supply curve model 143 has a supply slope model 143a and a supply intercept model 143b.

[0021] The supply slope model 143a is a machine learning model that inputs the contract price in the second time slot, which is time-series data, and outputs an estimated value of the slope of a supply approximation equation that approximates the supply curve in the first time slot for each second time slot to a linear function. For example, the supply slope model 143a is a model that uses an RNN. More specifically, the supply slope model 143a can use an LSTM network, which is a type of RNN.

[0022] The supply intercept model 143b is a machine learning model that inputs the contract price in the second time slot, which is time series data, and outputs an estimated value of an intercept of a supply approximation equation that approximates the supply curve in the first time slot for each second time slot to a linear function. For example, the supply intercept model 143b is a model that uses an RNN. More specifically, the supply intercept model 143b can use an LSTM network.

[0023] The contract price model 141 takes as input information on the slope and intercept corresponding to the demand curve in the first time slot, and outputs the contract price in the first time slot. That is, the contract price model 141 takes as input the values ​​of the slope and intercept of the demand approximation equation estimated by the demand curve model 142, and the values ​​of the slope and intercept of the supply approximation equation estimated by the supply curve model 143, and outputs the contract price in the first time slot. When the contract price in the second time slot is input, the contract price model 141 performs learning using the output value when the values ​​of the slope and intercept estimated by the demand curve model 142 and the supply curve model 143 are input, and the contract price in the first time slot.

[0024] In the learning phase, the approximation unit 12 approximates a known demand curve and supply curve to obtain a demand approximation equation and a supply approximation equation that become learning data. The approximation unit 12 has an initial function generation unit 21, a smoothing unit 122, and an intercept calculation unit 123. For example, the demand approximation equation is expressed by the following formula (1), and the supply approximation equation is expressed by the following formula (2).

[0025]

number

number

[0026] Here, a Buy is the slope of the demand approximation equation, and b Buy is the intercept of the demand approximation equation. Also, a Sell is the slope of the supply approximation equation, and b Sell is the intercept of the supply approximation equation. That is, a Buy and b Buy The demand approximation formula is defined as Sell and b Sell The supply approximation formula is defined as follows:

[0027] The initial function generating unit 121 generates an initial function, which is a provisionally determined basic linear function for approximating the demand curve and the supply curve to a linear function in order to obtain learning data. FIG. 2 is a diagram for explaining the generation of the initial function. The initial function generating unit 121 acquires information on the demand curve 201 and the supply curve 202 in the first time period shown in FIG. 2. Next, the initial function generating unit 121 identifies an intersection 203 of the demand curve 201 and the supply curve 202.

[0028] Next, the initial function generating unit 121 extracts an appropriate point on the demand curve as an extraction point. In this embodiment, the initial function generating unit 121 sets the bid price for the bid amount immediately before the bid price drops most rapidly as the extraction point. In other words, the initial function generating unit 121 extracts the maximum change rate starting point 204, which is the starting point of the place where the rate of decline is the highest on the demand curve 201. Next, the initial function generating unit 121 creates a demand initial function 211, which is a linear function that passes through the intersection point 203 and the maximum change rate starting point 204.

[0029] Next, the initial function generating unit 121 extracts an appropriate point on the supply line as an extraction point for the supply curve as well. In this embodiment, the initial function generating unit 121 sets the bid price for the bid amount immediately before the bid price first increases as the extraction point. In other words, the initial function generating unit 121 extracts an increase start point 205, which is a point on the supply curve 202 where the bid price starts to increase from 0. Next, the initial function generating unit 121 creates a supply initial function 212, which is a linear function that passes through the intersection point 203 and the increase start point 205.

[0030] The initial function generating unit 121 similarly generates a demand initial function and a supply initial function for each of a plurality of first time periods used for learning, which are a plurality of different time periods used as data. After that, the initial function generating unit 121 outputs information on the slope and intercept of each of the generated demand initial functions, and information on the slope and intercept of each of the generated supply initial functions to the smoothing unit 122. This demand initial function is an example of a "first function", and the supply initial function is an example of a "second function".

[0031] Fig. 3 is a diagram showing an example of time-series changes in the slope and intercept values ​​of the initial demand function created by the initial function generation unit. Graph 221 in Fig. 3 shows the time-series changes in the slope of the initial demand function, and graph 222 shows the time-series changes in the slope of the initial demand function. Graph 221 shows the measurement unit time slot over time on the horizontal axis, and the value of the slope on the vertical axis. Graph 222 shows the measurement unit time slot over time on the horizontal axis, and the value of the intercept on the vertical axis.

[0032] As shown in graphs 221 and 222, the slope and intercept of the demand initial function change frequently and significantly along the time series. Therefore, if the demand curve model 142 is trained using the data of the slope and intercept shown in graphs 221 and 222, it may be difficult to grasp the trend and the estimation accuracy of the slope and intercept may decrease. Therefore, it is not preferable to train the demand curve model 142 using the demand initial function as it is. Therefore, the approximation unit 12 performs the following process.

[0033] The smoothing unit 122 receives information on the slope and intercept of the demand initial function 211 and information on the slope and intercept of the supply initial function 212 for each of a plurality of different first time slots from the initial function generating unit 121. Then, the smoothing unit 122 smoothes the value of the slope of the demand initial function 211 along the time series so that the change becomes smooth, that is, so as to suppress the frequency of large fluctuations. Specifically, the smoothing unit 122 performs smoothing by calculating a moving average of the value of the slope of the demand initial function 211 along the time series. For example, the smoothing unit 122 smoothes the slope using an arithmetic average of the values ​​of seven consecutive measurement unit time slots along the time series.

[0034] Fig. 4 is a diagram showing an example of time-series changes in the slope and intercept values ​​after smoothing. For example, smoothing unit 122 calculates a slope value in which changes such as those shown in graph 231 in Fig. 4 have been smoothed by smoothing the slope information shown in graph 221 in Fig. 3 using a moving average. Graph 231 shows the measurement unit time period over time on the horizontal axis, and the slope value on the vertical axis. As shown in graph 231, the change in the slope value after smoothing is smoother, with the frequency of fluctuations being suppressed compared to graph 221.

[0035] In this embodiment, the smoothing unit 122 uses a moving average as a noise removal filter for smoothing the change in the slope value along the time series, but it is also possible to use other noise removal filters. For example, the smoothing unit 122 can perform smoothing using an exponential smoothing filter or a Karma filter.

[0036] The smoothing unit 122 smoothes the value of the slope of the supply initial function 212 along the time series so that the change is smooth, similar to the case of the supply initial function 211. Hereinafter, the slope obtained by smoothing the slopes of the demand initial function and the supply initial function is referred to as the "slope after smoothing". After that, the smoothing unit 122 outputs the value of the slope of the demand initial function and the supply initial function after smoothing for each of the multiple different first time slots to the intercept calculation unit 123.

[0037] The intercept calculation unit 123 receives an input of the smoothed slope value for each of the demand initial function and the supply initial function for each of the different first time slots from the smoothing unit 122. Fig. 5 is a diagram for explaining the creation of the demand approximation equation and the supply approximation equation.

[0038] Next, the intercept calculation unit 123 calculates, for each of a plurality of different first time slots, a demand approximation equation 241 that is a linear function having a slope after smoothing for the demand initial function and passing through the intersection point 203. In this case, since the slope has been smoothed, as shown in FIG. 5, most of the demand approximation equations 241 do not pass through the maximum change rate starting point 204. Then, the intercept calculation unit 123 obtains the intercept of the calculated demand approximation equation 241.

[0039] In the same manner, intercept calculation unit 123 calculates, for each of a plurality of different first time slots, supply approximation equation 242, which is a linear function having a slope after smoothing for the supply initial function and passing through intersection point 203. In this case as well, by smoothing the slope, many of supply approximation equations 242 do not pass through rise start point 205, as shown in Fig. 5. Then, intercept calculation unit 123 calculates the intercept of the calculated supply approximation equation 242.

[0040] For example, the change over time of the intercept calculated by intercept calculation unit 123 is as shown in graph 232 in Fig. 4. In graph 232, the horizontal axis indicates the measurement unit time period over time, and the vertical axis indicates the intercept value. As shown in graph 232, the change in the intercept value calculated by intercept calculation unit 123 is smoother, with less frequent fluctuations, compared to graph 222 in Fig. 3.

[0041] In this way, by smoothing the temporal changes in the slope and intercept, when the demand curve model 142 is made to learn using this information, appropriate learning can be performed, and the estimation accuracy of the slope and intercept can be improved. In addition, in this embodiment, the change in the intercept value is also smoothed by smoothing the change in the slope, but the present invention is not limited to this, and the smoothing unit 122 may simultaneously smooth the changes in the slope and intercept of the demand curve.

[0042] For example, when smoothing the change in the slope to smooth the change in the intercept value, the demand approximation formula 241 and the supply approximation formula 242 may be determined by the following process. The initial function generation unit 121 extracts the slope values ​​of the demand initial function 211 and the supply initial function 212 without completely creating them. Next, the smoothing unit 122 receives the slope values ​​extracted by the initial function generation unit 121 and smoothes the slope values. The intercept calculation unit 123 calculates the intercept values ​​from the smoothed slopes. In this manner, the demand approximation formula 241 and the supply approximation formula 242 can be determined.

[0043] In this way, the approximation unit 12 performs a process of approximating the demand curve and the supply curve to the demand approximation equation and the supply approximation equation in each of a plurality of different first time periods. In this way, the approximation unit 12 can create a plurality of data to be used for learning the demand curve model 142 and the supply curve model 143. In this way, the first time period for which the demand approximation equation and the supply approximation equation to be used as learning data are obtained is an example of a "first period."

[0044] Returning to FIG. 1, the description will be continued. The model generation unit 13 has a first model generation unit 131 and a second model generation unit 132. The model generation unit 13 trains a demand curve model 142 using information on the slope and intercept of the demand approximation equation in a plurality of first time periods obtained by the approximation unit 12 and information on the contract price in a second time period corresponding to each of the first time periods. The model generation unit 13 also trains a supply curve model 143 using information on the slope and intercept of the supply approximation equation in a plurality of first time periods obtained by the approximation unit 12 and information on the contract price in a second time period corresponding to each of the first time periods. The details of the training will be described below.

[0045] That is, the model generation unit 13 performs learning using data on the slopes and intercepts of the demand approximation equation, which is the first approximation equation, and the supply approximation equation, which is the second approximation equation, to generate the demand curve model 142 and the supply curve model 143, which are machine learning models. More specifically, the model generation unit 13 performs learning using data on a plurality of slopes and a plurality of intercepts corresponding to each of the plurality of demand approximation equations and the plurality of supply approximation equations.

[0046] The first model generation unit 131 is in charge of training the demand curve model 142. The first model generation unit 131 acquires information on the slopes of the demand approximation equation for each of the different first hours from the smoothing unit 122. In addition, the first model generation unit 131 acquires information on the intercepts of the demand approximation equation for each of the different first hours from the smoothing unit 122.

[0047] Next, the first model generation unit 131 acquires the contract price of the second time slot corresponding to each of the first time slots for which the demand approximation equation has been obtained from the data storage unit 11. In addition, the first model generation unit 131 acquires, for example, information on the temperature, the day of the week, and the power usage rate for each of the first time slots from the data storage unit 11.

[0048] Then, the first model generation unit 131 trains the demand curve model 142 using, for example, the contract price for the second time period, the temperature, day of the week, and power usage rate for the first time period, and the slope and intercept of the corresponding demand approximation equation for each first time period as learning data. More specifically, the first model generation unit 131 inputs, for example, the contract price for the second time period, the temperature, day of the week, and power usage rate for the first time period to the demand slope model 142a. Then, the first model generation unit 131 adjusts the parameters of the demand slope model 142a based on the error between the output from the demand slope model 142a and the slope of the corresponding demand approximation equation for each first time period. Also, the first model generation unit 131 inputs, for example, the contract price for the second time period, the temperature, day of the week, and power usage rate for the first time period to the demand intercept model 142b. Then, the first model generation unit 131 adjusts the parameters of the demand intercept model 142b based on the error between the output from the demand intercept model 142b and the intercept of the corresponding first hourly demand approximation equation.

[0049] The second model generation unit 132 is responsible for training the supply curve model 143. The second model generation unit 132 acquires information on the slopes of the supply approximation equation for each of a plurality of different first times from the smoothing unit 122. In addition, the second model generation unit 132 acquires information on the intercepts of the supply approximation equation for each of a plurality of different first times from the smoothing unit 122.

[0050] Next, the second model generation unit 132 acquires the contract price of the second time slot corresponding to each of the first time slots for which the supply approximation equation has been obtained from the data storage unit 11. In addition, the second model generation unit 132 acquires, for example, information on the temperature, the day of the week, and the power usage rate in each of the first time slots from the data storage unit 11.

[0051] The second model generation unit 132 trains the supply curve model 143 using, for example, the contract price for the second time period, the temperature, day of the week, and power usage rate for the first time period, and the slope and intercept of the corresponding supply approximation equation for each first time period as learning data. More specifically, the second model generation unit 132 inputs, for example, the contract price for the second time period, the temperature, day of the week, and power usage rate for the first time period to the supply slope model 143a. The second model generation unit 132 adjusts the parameters of the supply slope model 143a based on the error between the output from the supply slope model 143a and the slope of the corresponding supply approximation equation for each first time period. The second model generation unit 132 also inputs, for example, the contract price for the second time period, the temperature, day of the week, and power usage rate for the first time period to the supply intercept model 143b. Then, the second model generating unit 132 adjusts the parameters of the supply intercept model 143b based on the error between the output from the supply intercept model 143b and the corresponding intercept of the first supply approximation equation for each time period.

[0052] Here, the demand for electricity is affected by, for example, temperature, day of the week, and power usage rate. Therefore, as described above, the first model generation unit 131 reflects information such as temperature and day of the week in the estimation of the slope and intercept of the demand approximation equation. In addition, since the supply amount of electricity varies depending on the demand amount, the supply amount is also affected by, for example, temperature, day of the week, or power usage rate. Therefore, as described above, the second model generation unit 132 reflects information such as temperature and day of the week in the estimation of the slope and intercept of the supply approximation equation. This allows the first model generation unit 131 and the second model generation unit 132 to predict the demand approximation equation and the supply approximation equation with high accuracy, taking into account the influence of temperature, day of the week, or power usage rate.

[0053] Also, when the contract volume approaches the total selling bid volume, the contract price rises sharply. Therefore, the first model generation unit 131 and the second model generation unit 132 may reflect information on the difference between the total selling bid volume and the contract volume in determining either or both of the slope and the intercept. For example, the first model generation unit 131 inputs learning data including the minimum value of the difference between the total selling bid volume and the contract volume among other trading time slots on the same day as the first time slot, and causes the demand slope model 142a and the demand intercept model 142b to estimate the slope and intercept of the demand approximation formula. As a result, the first model generation unit 131 obtains the slope and intercept of the demand approximation formula that takes into account the difference between the total selling bid volume and the contract volume, and by causing the demand slope model 142a and the demand intercept model 142b to learn using these, training with improved estimation accuracy can be performed. By predicting the contract price using the slope and intercept of the demand approximation equation obtained from the demand slope model 142a and the demand intercept model 142b that have been trained in this way, a highly accurate contract price can be obtained. This is also true for the second model generation unit 132.

[0054] In addition, in this embodiment, the model generation unit 13 trains both the demand curve model 142 and the supply curve model 143 using a demand approximation formula and a supply approximation formula in which the demand curve and the supply curve are approximated to a linear function. However, for example, when the demand curve is approximated by a sigmoid function, the approximation function has a large amount of change over time, making it difficult to use it to estimate the contract price. On the other hand, when the supply curve is approximated by a sigmoid function, the approximation function does not have a large amount of change over time, making it possible to use it to estimate the contract price. Therefore, the model generation unit 13 may train the demand curve model 142 to estimate a demand approximation formula in which the demand curve is approximated to a linear function, and train the supply curve model 143 to estimate a demand curve approximated by a sigmoid function.

[0055] The estimation unit 15 estimates the contract price in the first time slot based on the input contract price in the second time slot, using the learned contract price model 141, the demand curve model 142, and the supply curve model 143. The details of the estimation unit 15 are described below.

[0056] The estimation unit 15 includes a demand curve estimation unit 151, a supply curve estimation unit 152, and an agreement price estimation unit 153, as shown in FIG. 1. Hereinafter, the first time period to be estimated is referred to as the "estimated time period", and the second time period for the estimated time period, i.e., the second time period from which the agreement price used in the estimation is obtained, is referred to as the "original estimation time period". Here, the estimated time period is a future time period relative to the time of estimation, and the original estimation time period is a past time period relative to the time of estimation. The user acquires an agreement price in the original estimation time period that has already been determined in the spot market. Then, the user inputs the agreement price in the original estimation time period that has already been determined in the spot market to the agreement price estimation device 1 using the user terminal 2, and requests estimation of the agreement price in the estimated time period.

[0057] The demand curve estimation unit 151 receives an input of a contract price in the estimation source time slot already determined in the spot market from the user terminal 2. Next, the demand curve estimation unit 151 inputs the contract price in the estimation source time slot to the learned demand slope model 142a. Then, the demand curve estimation unit 151 acquires an estimation result of the slope of the demand approximation equation in the estimation target time slot output from the demand slope model 142a. In addition, the demand curve estimation unit 151 inputs the contract price in the estimation source time slot to the learned demand intercept model 142b. Then, the demand curve estimation unit 151 acquires an estimation result of the intercept of the demand approximation equation in the estimation target time slot output from the demand intercept model 142b. Then, the demand curve estimation unit 151 outputs information on the slope and intercept of the demand approximation equation in the estimation target time slot to the contract price estimation unit 153 as information indicating an estimated demand curve in the estimation target time slot.

[0058] The supply curve estimation unit 152 receives an input of a contract price in the estimation source time slot already determined in the spot market from the user terminal 2. Next, the supply curve estimation unit 152 inputs the contract price in the estimation source time slot to the learned supply slope model 143a. Then, the supply curve estimation unit 152 acquires an estimation result of the slope of the supply approximation equation in the estimation target time slot output from the supply slope model 143a. In addition, the supply curve estimation unit 152 inputs the contract price in the estimation source time slot to the learned supply intercept model 143b. Then, the supply curve estimation unit 152 acquires an estimation result of the intercept of the supply approximation equation in the estimation target time slot output from the supply intercept model 143b. Then, the supply curve estimation unit 152 outputs information on the slope and intercept of the supply approximation equation in the estimation target time slot to the contract price estimation unit 153 as information indicating an estimated demand curve in the estimation target time slot.

[0059] Since the characteristics of the supply and demand curve and the supply curve appear in the slopes and intercepts of the demand approximation equation and the supply approximation equation, the contract price estimation unit 153 predicts the contract price using these slopes and intercepts as explanatory variables of the contract price model 141. The operation of the contract price estimation unit 153 will be described below.

[0060] The contract price estimation unit 153 may train the contract price model 141 in the learning phase. For example, the contract price estimation unit 153 acquires information on the slope and intercept of each of the demand approximation equation and the supply approximation equation in a plurality of first time periods estimated by the demand curve model 142 and the supply curve model 143 from past data. In addition, the contract price estimation unit 153 acquires the actual contract price for each first time period from which the slope and intercept of the demand approximation equation and the supply approximation equation are obtained. Then, the contract price estimation unit 153 can train the contract price model 141 using the slope and intercept of each of the demand approximation equation and the supply approximation equation in the first time period and the actual contract price in the first time period as learning data. However, the contract price estimation unit 153 may use the contract price model 141 trained by a unit other than the contract price estimation unit 153.

[0061] Also, in the estimation phase, the contract price estimation unit 153 receives input of information on the slope and intercept of each of the estimated demand approximation equation and supply approximation equation in the estimation target time zone from the demand curve estimation unit 151 and the supply curve estimation unit 152. Next, the contract price estimation unit 153 inputs information on the slope and intercept of each of the estimated demand approximation equation and supply approximation equation in the estimation target time zone to the contract price model 141. Then, the contract price estimation unit 153 obtains the estimated result of the contract price in the estimation target time zone output from the contract price model 141. Thereafter, the contract price estimation unit 153 transmits the estimated result of the contract price in the estimation target time zone to the user terminal 2 through the estimation result in response to the request from the user, and provides it to the user.

[0062] Fig. 6 is a diagram showing an outline of the estimation process by the estimation unit. LSTMs 301 and 302, and 311 and 312 in Fig. 6 are specific examples of the demand slope model 142a, the demand intercept model 142b, the supply slope model 143a, and the supply intercept model 143b, respectively. Also, LSTM 340 is a specific example of the contract price model 141. Here, the estimation process in the estimation phase will be collectively described with reference to Fig. 6.

[0063] The demand curve estimation unit 151 inputs the confirmed contract price in the estimation source time zone into the LSTM 301 corresponding to the demand slope model 142a, and obtains the slope 321 of the demand approximation equation to be output. In addition, the demand curve estimation unit 151 inputs the confirmed contract price in the estimation source time zone into the LSTM 302 corresponding to the demand intercept model 142b, and obtains the intercept 322 of the demand approximation equation to be output.

[0064] The supply curve estimation unit 152 inputs the confirmed contract price in the estimation source time zone into the LSTM 311 corresponding to the supply slope model 143a, and obtains the output slope 331 of the supply approximation equation. In addition, the supply curve estimation unit 152 inputs the confirmed contract price in the estimation source time zone into the LSTM 312 corresponding to the supply intercept model 143b, and obtains the output intercept 332 of the supply approximation equation.

[0065] The contract price estimation unit 153 inputs the slope 321 of the demand approximation equation, the intercept 322 of the demand approximation equation, the slope 331 of the supply approximation equation, and the intercept 332 of the supply approximation equation to the LSTM 340 corresponding to the contract price model 141, and obtains an estimated contract price value 350 in the estimated target time period to be output. In this way, the estimation unit 15 can estimate the contract price in the estimated time period.

[0066] FIG. 7 is a diagram showing a comparison of prediction errors between estimation by the contract price estimation device according to the first embodiment and estimation using LSTM alone. Here, estimation using LSTM alone refers to estimation using LSTM that has learned to estimate contract prices without using demand curves or supply curves. Here, the measurement unit is 30 minutes. Table 350 shows the results of prediction errors by sequential one-period-ahead prediction in which the second time slot, which is the estimation source time slot, is the measurement unit immediately before the first time slot, which is the estimation target time slot.

[0067] As shown in table 350, in the case of estimation by the contract price estimation device 1 according to this embodiment, the prediction error is 7.2867. In contrast, in the case of estimation by LSTM alone, the prediction error is 12.0809. In other words, the slopes and intercepts of the estimated demand approximation equation and supply approximation equation accurately reflect the characteristics of the demand curve and supply curve, and it can be said that the contract price estimation device 1 can keep the prediction error small by predicting the contract price using these slopes and intercepts as explanatory variables.

[0068] As described above, the estimation unit 15 uses the trained machine learning model to estimate the intersection between the demand curve and the supply curve in the second period corresponding to the first period used for learning. The value represented by the intersection between the demand curve and the supply curve corresponds to the contract price. More specifically, the estimation unit 15 uses the machine learning model to estimate the intercept and slope of each of the first approximation equation and the second approximation equation in the second time period. Then, the estimation unit 15 executes a process of estimating the intersection between the first curve and the second curve, including a process of estimating the intersection between the first curve and the second curve based on the estimated intercept and slope.

[0069] Here, in this embodiment, the demand curve model 142, the supply curve model 143, and the contract price model 141 are trained separately, but they can also be trained together. That is, for a given contract price, an estimated contract price obtained using the demand curve model 142, the supply curve model 143, and the contract price model 141 is obtained, and the parameters of the demand curve model 142, the supply curve model 143, and the contract price model 141 may be adjusted based on the error between the acquired contract price and the estimated result.

[0070] Fig. 8 is a flowchart of the learning process of the demand curve model and the supply curve model in the learning phase. Next, the flow of the learning process of the demand curve model 142 and the supply curve model 143 in the learning phase of the contract price estimation device 1 according to this embodiment will be described with reference to Fig. 8.

[0071] The initial function generating unit 121 acquires, for a plurality of past first time slots, information on demand curves and supply curves in the first time slots that have already been established and published in the spot market from the data storage unit 11 (step S101).

[0072] Next, the initial function generating unit 121 selects one first time slot from among the unselected first time slots in the plurality of first time slots for which the demand curve and the supply curve have been obtained (step S102).

[0073] Next, the initial function generating unit 121 identifies the maximum change rate start point of the demand curve in the selected first time slot (step S103).

[0074] Next, the initial function generating unit 121 identifies the rising start point of the supply curve in the selected first time slot (step S104).

[0075] Next, the initial function generating unit 121 creates a demand initial function, which is a linear function that connects the intersection point between the demand curve and the supply curve and the maximum change rate start point. The initial function generating unit 121 also creates a supply initial function, which is a linear function that connects the intersection point between the demand curve and the supply curve and the increase start point (step S105).

[0076] Next, the initial function generating unit 121 judges whether or not the creation of the demand initial function and the supply initial function is completed for all the first time periods for which the demand curve and the supply curve are acquired (step S106). If there are any first time periods for which the demand initial function and the supply initial function are not created (step S106: No), the initial function generating unit 121 returns to step S102.

[0077] On the other hand, when the creation of the demand initial function and the supply initial function for all the first time slots is completed (step S106: Yes), the initial function generation unit 121 outputs the demand initial function and the supply initial function for each of the first time slots to the smoothing unit 122. The smoothing unit 122 acquires the change over time in the slope of the demand initial function and the supply initial function created by the initial function generation unit 121. Then, the smoothing unit 122 smoothes the slope of each of the demand initial function and the supply initial function for each first time slot by using a moving average for the change in slope over time (step S107).

[0078] The intercept calculation unit 123 obtains the smoothed slopes of the initial demand function and the initial supply function for each first time slot from the smoothing unit 122. Next, the intercept calculation unit 123 selects one first time slot from among the unselected first time slots in the multiple first time slots for which the initial demand function and the initial supply function have been calculated (step S108).

[0079] Next, the intercept calculation unit 123 creates a demand approximation equation, which is a linear function that has a smoothed slope of the demand initial function and passes through the intersection of the demand curve and the supply curve, for the selected first time slot. Also, the intercept calculation unit 123 creates a supply approximation equation, which is a linear function that has a smoothed slope of the supply initial function and passes through the intersection of the demand curve and the supply curve, for the selected first time slot (step S109).

[0080] Next, the intercept calculation unit 123 judges whether or not the creation of the demand approximation formula and the supply approximation formula is completed for all the first time slots for which the demand initial function and the supply initial function have been obtained (step S110). If there are any first time slots for which the demand approximation formula and the supply approximation formula have not been created (step S110: No), the initial function generation unit 121 returns to step S108.

[0081] On the other hand, when the creation of the demand approximation equation and the supply approximation equation is completed for all the first time slots (step S110: Yes), the intercept calculation unit 123 identifies the intercepts of the demand approximation equation and the supply approximation equation for each first time slot. The first model generation unit 131 and the second model generation unit 132 acquire the smoothed slopes of the demand initial function and the supply initial function for each first time slot from the smoothing unit 122. In addition, the first model generation unit 131 and the second model generation unit 132 acquire the intercepts of the demand approximation equation and the supply approximation equation for each first time slot from the intercept calculation unit 123. Furthermore, the first model generation unit 131 and the second model generation unit 132 acquire the contract price already determined in the spot market from the data storage unit 11 for the second time slot corresponding to the first time slot for which the demand approximation equation and the supply approximation equation are estimated. Furthermore, in this embodiment, the first model generation unit 131 and the second model generation unit 132 acquire information on the temperature, the day of the week, and the power usage rate in the first time period from the data storage unit 11.

[0082] Then, the first model generation unit 131 causes the demand curve model 142 to learn using a data set including the contract price in the second time slot and the slope and intercept of the demand approximation equation in the first time slot as learning data (step S111). Here, in this embodiment, the first model generation unit 131 causes the temperature, day of the week, and power usage rate for each first time slot to be included in the data set to be the learning data. More specifically, the first model generation unit 131 causes each of the demand slope model 142a and the demand intercept model 142b to learn.

[0083] The second model generation unit 132 also causes the supply curve model 143 to learn using a data set including the contract price in the second time slot and the slope and intercept of the supply approximation equation in the first time slot as learning data (step S112). Here, in this embodiment, the second model generation unit 132 also causes the data set to be learning data to include, for example, the temperature, day of the week, and power usage rate for each first time slot. More specifically, the second model generation unit 132 causes each of the supply slope model 143a and the supply intercept model 143b to learn. With the above, learning of the demand curve model 142 and the supply curve model 143 in the learning phase is completed.

[0084] 9 is a flowchart of the estimation process by the contract price estimation device in the estimation phase. Next, the flow of the estimation process of the contract price in the estimation phase by the contract price estimation device 1 according to this embodiment will be described with reference to FIG.

[0085] The demand curve estimation unit 151 and the supply curve estimation unit 152 receive an input of the contract price for the second time slot, which is the estimation source time slot, from the user terminal 2 together with a request for estimation of the contract price for the estimation target time slot (step S201).

[0086] Next, the demand curve estimation unit 151 inputs the contract price in the estimation source time slot into the demand slope model 142a and the demand intercept model 142b, which are the learned demand curve model 142. Then, the demand curve estimation unit 151 obtains an estimate of the slope of the demand approximation equation in the first time slot, which is the estimation target time slot, from the demand slope model 142a. The demand curve estimation unit 151 also obtains an estimate of the intercept of the demand approximation equation in the estimation target time slot from the demand slope model 142a (step S202).

[0087] Furthermore, the supply curve estimation unit 152 inputs the contract price in the estimation source time slot into the supply slope model 143a and the supply intercept model 143b, which are the learned supply curve model 143. Then, the supply curve estimation unit 152 obtains an estimated value of the slope of the supply approximation equation in the estimation target time slot from the supply slope model 143a. Furthermore, the supply curve estimation unit 152 obtains an estimated value of the intercept of the supply approximation equation in the estimation target time slot from the supply slope model 143a (step S203).

[0088] The contract price estimation unit 153 inputs the estimated values ​​of the slope and intercept of each of the demand approximation equation and the supply approximation equation in the estimation target time period to the contract price model 141. Then, the contract price estimation unit 153 obtains the estimated value of the contract price in the estimation target time period output from the contract price model 141 (step S204).

[0089] Thereafter, the contract price estimation unit 153 transmits an estimated value of the contract price in the estimation target time period to the user terminal 2 as an estimation result in response to the request from the user, and provides it to the user (step S205).

[0090] (Hardware configuration) 10 is a hardware configuration diagram of the contract price estimation device according to the embodiment 1. Next, an example of a hardware configuration for realizing each function of the contract price estimation device 1 will be described with reference to FIG.

[0091] 10, the contract price estimation device 1 includes, for example, a CPU (Central Processing Unit) 91, a memory 92, a hard disk 93, and a network interface 94. The CPU 91 is connected to the memory 92, the hard disk 93, and the network interface 94 via a bus.

[0092] The network interface 94 is an interface for communication between the contract price estimation device 1 and an external device. The network interface 94 relays communication between the user terminal 2 and the CPU 91, for example.

[0093] The hard disk 93 is an auxiliary storage device. The hard disk 93 realizes the functions of the data storage unit 11 and the model storage unit 14 illustrated in Fig. 1. The hard disk 93 also stores various programs including programs for realizing the functions of the approximation unit 12, the model generation unit 13, and the estimation unit 15 illustrated in Fig. 1.

[0094] The memory 92 is a main storage device, and may be, for example, a dynamic random access memory (DRAM).

[0095] The CPU 91 reads out various programs from the hard disk 93, and loads and executes them in the memory 92. In this way, the CPU 91 realizes the functions of the approximation unit 12, the model generation unit 13, and the estimation unit 15 illustrated in FIG.

[0096] As described above, the contract price estimation device according to the present embodiment approximates the demand curve and the supply curve in the first time period to a linear function, and further smooths the slope and intercept along the time series to create the demand approximation formula and the supply approximation formula. Then, the contract price estimation device trains the demand curve model and the supply curve model using a data set including the slope and intercept of the created demand approximation formula and the supply approximation formula and the contract price in the second time period as learning data. In this way, by smoothing the fluctuation along the time series of the demand approximation formula and the supply approximation formula, the parameters of the demand curve model and the supply curve model can be properly trained. Then, by properly training the parameters, the demand curve model and the supply curve model that can accurately estimate the demand approximation formula and the supply approximation formula can be generated. Note that the above method is not limited to the demand curve and the supply curve, and can be adopted for other types of curves as long as they are similar curves, and it is possible to determine the curve model and the intersection point with high accuracy.

[0097] Furthermore, the contract price estimation device estimates the contract price in the estimation target time period using the learned demand curve model and supply curve model, and the contract price model that estimates the contract price from the estimated values ​​of the slope and intercept. In other words, by improving the estimation accuracy of the demand approximation equation and the supply approximation equation by the demand curve model and the supply curve model, the prediction accuracy of the contract price can be improved, and it becomes possible to predict the spot price with high accuracy using the supply and demand curve data. EXAMPLES

[0098] 11 is a diagram illustrating an example of a configuration of a computer system according to Example 2. A computer system 10 according to Example 2 is a computer system that generates a model for estimating a demand curve and a supply curve in a spot market for electricity trading with high accuracy and predicts a contract price using the model.

[0099] In the computer system 10, a plurality of bidding computers 100, 100a, . . ., a trading system 200, a power information server 300, and a weather information server 400 are connected to a network 20.

[0100] The multiple bidding computers 100 and 100a are, for example, computers used for bidding by the power generation company 31 or the power retailer 32. For example, the power generation company 31 uses the bidding computer 100, and the retailer 32 uses the bidding computer 100a.

[0101] The trading system 200 is a computer that determines an agreement price for each delivery time slot, which is a time slot of the length of a measurement unit, and executes an agreement between selling orders and buying orders bid within each delivery time slot. The trading system 200 acquires selling orders for electricity from a bidding computer 100 used by, for example, a power generation company 31, and acquires buying orders for electricity from a bidding computer 100a used by a retail company 32. The trading system 200 determines a supply curve and a demand curve based on the selling orders and buying orders, and determines the intersection of these curves as the contract amount and the contract price.

[0102] The power information server 300 is a computer that provides data on the power usage status. For example, the power information server 300 provides power information indicating the power usage rate for each date in response to a request from the bidding computers 100 and 100a.

[0103] The weather information server 400 is a computer that provides data related to weather information. For example, the weather information server 400 provides weather data indicating temperatures for each date in response to requests from the bidding computers 100 and 100a.

[0104] Spot trading of electricity is carried out by such a computer system 10. For example, in the case of a one-day-ahead market, the day before the day on which electricity is bought and sold (delivery date) is the trading day, and the contract price is determined and the transaction is concluded on the trading day.

[0105] In spot trading of electricity, if the power generation company 31 can accurately estimate the contract price for each delivery time slot on the delivery day before the trading day, the power generation company 31 can sell the generated electricity at a high price by placing a sell order in the delivery time slot when the contract price is high. Also, if the retail business operator 32 can accurately estimate the contract price for each delivery time slot on the delivery day before the trading day, the retail business operator 32 can purchase electricity at a low price by placing a buy order in the delivery time slot when the contract price is low. Therefore, the bidding computers 100 and 100a predict the contract price using the function of the contract price estimation device 1.

[0106] Here, the estimation of the contract price will be explained using a bidding computer 100 as an example. FIG. 12 is a block diagram of the bidding computer. The bidding computer 100 has the functions of the contract price estimation device 1 shown in FIG. 1. That is, the bidding computer 100 has the data storage unit 11, approximation unit 12, model generation unit 13, model storage unit 14, and estimation unit 15 shown in FIG. 1. Although detailed functions of each unit are omitted in FIG. 12, in reality, each unit has the same function as in FIG. 1. The bidding computer 100 also has a bidding unit 101, an input device 102, and an output device 103.

[0107] The input device 102 is a keyboard, a mouse, etc. The output device 103 is a monitor, etc.

[0108] The bidding unit 101 receives instructions from a user via the input device 102, transmits information on a selling bid or a buying bid to the trading system 200, and places a bid in the electricity trading market. The bidding unit 101 also obtains transaction data on the concluded transaction from the trading system 200, and stores it in the data storage unit 11.

[0109] The data storage unit 11 stores transaction data, power data, and weather data. The transaction data is data related to confirmed transactions for each delivery time period. The power data is information such as power usage rate. The weather data is data related to weather such as temperature.

[0110] The data storage unit 11 also acquires and stores power data 112 relating to the power usage rate and the like from the power information server 300. The data storage unit 11 also acquires and stores weather data 113 including information such as temperature from the weather information server 400.

[0111] The initial function generating unit 121 of the approximation unit 12 acquires from the data storage unit 11 the demand curve and the supply curve for each of a plurality of delivery time slots for which transactions have already been contracted. Then, the initial function generating unit 121 of the approximation unit 12 generates an initial demand function and an initial supply function from the demand curve and the supply curve. The smoothing unit 122 of the approximation unit 12 smoothes the change in slope over time of the initial demand function and the initial supply function using a moving average or the like. The intercept calculation unit 123 of the approximation unit 12 calculates a smoothed intercept using the smoothed slope. In this manner, the approximation unit 12 creates a demand approximation equation and a supply approximation equation.

[0112] The first model generation unit 131 and the second model generation unit 132 of the model generation unit 13 acquire the contract price, temperature, day of the week, and power usage rate for the first time slot and the second time slot from the data storage unit 11. Then, the first model generation unit 131 and the second model generation unit 132 of the model generation unit 13 generate a data set including these, and use it as learning data.

[0113] Next, the first model generation unit 131 of the model generation unit 13 causes the demand slope model 142a and the demand intercept model 142b included in the demand curve model 142 to learn using the learning data. Also, the second model generation unit 132 of the model generation unit 13 causes the supply slope model 143a and the supply intercept model 143b included in the supply curve model 143 to learn using the learning data.

[0114] The demand curve estimation unit 151 of the estimation unit 15 acquires the contract price, temperature, day of the week, and power usage rate in the estimation source time slot input using the input device 102. Then, the demand curve estimation unit 151 of the estimation unit 15 inputs the contract price, temperature, day of the week, and power usage rate in the estimation source time slot to the learned demand curve model 142 and supply curve model 143. Then, the demand curve estimation unit 151 of the estimation unit 15 acquires the estimated values ​​of the slope and intercept of each of the demand approximation equation and the supply approximation equation in the estimation source time slot. Next, the demand curve estimation unit 151 of the estimation unit 15 inputs the estimated values ​​of the slope and intercept of each of the demand approximation equation and the supply approximation equation in the estimation source time slot to the supply curve model 143 to acquire the estimated value of the contract price in the estimation target time slot. Then, the estimation unit 15 transmits the estimated result of the contract price in the estimation target time slot to the output device 103 to output it. A user can use the bidding unit 101 to make a bid based on the estimated value of the contract price in the estimation target time period outputted to the output device 103 .

[0115] For example, the estimation unit 15 estimates the contract price by setting the first time slot one day after the second time slot, which is the estimation source time slot, as the estimated time slot. This allows the user to obtain an estimate of the contract price one day in advance and to make an appropriate bid in the relevant time slot one day later.

[0116] As described above, the bidding computer in the computer system according to this embodiment uses the contract price, temperature, day of the week, and power usage rate to determine a demand approximation equation and a supply approximation equation that approximate the demand curve and the supply curve to a linear function. The bidding computer then causes the demand curve model and the supply curve model to learn using the demand approximation equation and the supply approximation equation. Thereafter, the bidding computer estimates the contract price using the learned demand curve model and supply curve model, and the contract price model. This improves the estimation accuracy of the demand approximation equation and the supply approximation equation, making it possible to predict the contract price with high accuracy.

[0117] Although the computer system 10 of the second embodiment predicts the contract price in the electricity trading market, the present invention is not limited to this. In other words, for transactions in markets other than electricity, it is possible to predict the contract price with high accuracy by obtaining an approximation of the linear function of the demand curve and the supply curve and having the demand curve model and the supply curve model learn the approximation. [Explanation of symbols]

[0118] 1. Contract price estimation device 2. User terminal 11 Data storage section 12 Approximation part 13 Model Generation Unit 14 Model Storage 15 Estimation part 121 Initial Function Generator 122 Smoothing section 123 Intercept calculation part 131 First model generation unit 132 Second Model Generation Unit 141 Transaction Price Model 142 Demand Curve Model 142a Demand Slope Model 142b Demand intercept model 143 Supply Curve Model 143a Supply Slope Model 143b Supply Intercept Model 151 Demand Curve Estimator 152 Supply curve estimation part 153 Contract Price Estimation Department

Claims

1. approximating a first curve and a second curve having one intersection with the first curve during a first period to a first approximation formula and a second approximation formula, which are linear functions passing through the intersection; generating a machine learning model by performing learning using data on the slope and intercept of the first approximation formula and the second approximation formula, respectively; Using the trained machine learning model, an intersection point between the first curve and the second curve in a second period corresponding to the first period is estimated. An estimation program that causes a computer to execute a process.

2. the approximation process includes a process of approximating the first curve and the second curve to a first approximation formula and a second approximation formula in each of a plurality of different first periods, The process of generating the machine learning model includes a process of performing learning using data of a plurality of slopes and a plurality of intercepts corresponding to each of the plurality of first approximation expressions and the plurality of second approximation expressions.

2. The estimation program according to claim 1 .

3. The approximation process is creating a first function connecting points on the first curve and points of intersection; creating a second function connecting points on the second curve and points of intersection; smoothing fluctuations in the slopes of the first function and the second function; A function having the smoothed slope of the first function and passing through an intersection point is defined as the first approximation formula; A function having the smoothed slope of the second function and passing through the intersection point is defined as the second approximation formula.

2. The estimation program according to claim 1, further comprising a process for:

4. The estimation program according to claim 1, characterized in that the process of estimating the intersection point between the first curve and the second curve includes a process of using the machine learning model to estimate an intercept and a slope of each of the first approximation equation and the second approximation equation in the second period, and estimating the intersection point between the first curve and the second curve based on the estimated intercept and slope.

5. the first curve is a demand curve; The second curve is the supply curve.

2. The estimation program according to claim 1 .

6. the first time period and the second time period are 30 minute periods; The second period of time is one day after the first period of time.

2. The estimation program according to claim 1 .

7. An information processing device, approximating a first curve and a second curve having one intersection with the first curve during a first period to a first approximation formula and a second approximation formula, which are linear functions passing through the intersection; generating a machine learning model by performing learning using data on the slope and intercept of the first approximation formula and the second approximation formula, respectively; Using the trained machine learning model, an intersection point between the first curve and the second curve in a second period corresponding to the first period is estimated. The estimation method according to claim 1,

8. an approximation unit that approximates a first curve and a second curve having one intersection with the first curve during a first period to a first approximation formula and a second approximation formula that are linear functions passing through the intersection; a model generation unit that generates a machine learning model by performing learning using data on the slopes and intercepts of the first approximation formula and the second approximation formula; an estimation unit that estimates an intersection point between the first curve and the second curve in a second period corresponding to a period after the first period by using the trained machine learning model; An information processing device comprising:

Images