Solar power output prediction device, power grid control system, supply and demand control system, learning device, solar power output prediction method, and solar power output prediction program

Abstract

To obtain a solar power output prediction device that can predict solar power output with a temporal resolution higher than the temporal resolution of the acquired solar radiation intensity prediction values. [Solution] The solar power output prediction device includes: a data acquisition unit that acquires one or more solar radiation intensity prediction values ​​from among the predicted solar radiation intensity values ​​for times from sunrise to sunset, and actual values ​​of values ​​indicating solar power output for time periods shorter than the time interval of the time series data; a model generation unit that calculates statistical quantities of the actual values ​​within a time period and generates a trained model that predicts statistical quantities of values ​​indicating solar power output in a time period using training data including the first solar radiation intensity prediction value and the month and the corresponding statistical quantities; and a prediction unit that inputs the second solar radiation intensity prediction value and the month for the day to be predicted into the trained model and predicts statistical quantities of values ​​indicating solar power output for each time period. The actual values ​​include one or more data corresponding to one or more times for each time period, and the time corresponding to the predicted value to be used includes a time within the time period.

Description

Technical Field

[0001] The present disclosure relates to a solar power generation output prediction device, a power system control system, a supply-demand control system, a learning device, a solar power generation output prediction method, and a solar power generation output prediction program.

Background Art

[0002] In recent years, the importance of expanding the use of renewable energy has been increasing, and the number of consumers installing distributed power sources such as solar power generation facilities and supplying power to the power system (transmission and distribution system) has been increasing. On the other hand, solar power generation facilities are difficult to arbitrarily adjust their power generation output like thermal power generation facilities, and since the power generation output varies with the variation of solar radiation intensity (sunlight), it is necessary to accurately predict the variation of the power generation output. For this reason, technical developments required to predict the power generation output of solar power generation facilities installed by each consumer are being carried out everywhere. For example, Patent Document 1 discloses a technique for predicting the power generation output of a solar power generation facility using a predicted value of solar radiation intensity.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In the technology described in Patent Document 1, the temporal resolution of the solar power generation output depends on the temporal resolution of the predicted solar radiation intensity obtained as meteorological information. On the other hand, in applications such as power grid operation (particularly power grid control), it may be necessary to predict the solar power generation output with a temporal resolution higher than that of the predicted solar radiation intensity. In such cases, the technology described in Patent Document 1 may not be usable in power grid control because the temporal resolution of the predicted solar power generation output does not meet the requirements. Therefore, it is desirable to predict solar power generation output with a temporal resolution higher than that of the acquired predicted solar radiation intensity.

[0005] This disclosure is made in view of the above, and aims to provide a solar power output prediction device that can predict solar power output with a temporal resolution higher than the temporal resolution of the acquired solar radiation intensity prediction value. [Means for solving the problem]

[0006] To solve the above-mentioned problems and achieve the objective, the solar power output forecasting device according to this disclosure includes a data acquisition unit that acquires solar radiation intensity forecast values, which are one or more data points corresponding to one or more times within the period from sunrise to sunset from the predicted solar radiation intensity values ​​obtained as time series data, and actual values ​​that indicate solar power output for time periods shorter than the time interval of the time series data, and calculates statistical quantities of the actual values ​​for each time period, and uses a first solar radiation intensity forecast value, which is a solar radiation intensity forecast value, and the month corresponding to the first solar radiation intensity forecast value as input data, and uses training data including the input data and the statistical quantities corresponding to the input data to perform machine learning. The system comprises a model generation unit that generates a trained model for predicting statistical quantities of values ​​indicating solar power generation output for each time period from predicted solar radiation intensity values ​​and months, and a prediction unit that inputs a second predicted solar radiation intensity value, which is the predicted solar radiation intensity value for the day to be predicted, and the months corresponding to the second predicted solar radiation intensity value, obtained by the data acquisition unit, into the trained model to predict statistical quantities of values ​​indicating solar power generation output for each time period as the output of the trained model. The actual values ​​include one or more data points corresponding to one or more times within each time period, and the times corresponding to the predicted solar radiation intensity values ​​used by the model generation unit and the prediction unit include times within the corresponding time period. [Effects of the Invention]

[0007] According to this disclosure, it is possible to predict solar power output with a temporal resolution higher than that of the acquired predicted solar radiation intensity values. [Brief explanation of the drawing]

[0008] [Figure 1] This figure shows an example configuration of a power grid control system including a solar power output prediction device according to an embodiment. [Figure 2] A flowchart illustrating an example of the procedure for generating a trained model in a solar power output prediction device. [Figure 3] A diagram showing an example of a section for externally predicted values. [Figure 4] This figure shows an example of external prediction values ​​used to generate a trained model. [Figure 5] A schematic diagram showing an example of time-series data of externally predicted values ​​as a graph. [Figure 6] A diagram showing an example of actual solar radiation intensity values ​​and the maximum and minimum values ​​within a time window. [Figure 7] A schematic diagram showing an example of a neural network. [Figure 8] A flowchart illustrating an example of the procedure for predicting statistics of solar radiation intensity. [Figure 9] This figure shows an example of the predicted external values ​​and the predicted maximum and minimum values ​​of solar radiation intensity for the day being predicted. [Figure 10] This diagram shows an example configuration of a computer system that implements a solar power generation output prediction device. [Figure 11] A diagram showing another example configuration of a power grid control system. [Figure 12] This diagram shows an example configuration of a supply and demand control system that includes a solar power output forecasting device. [Modes for carrying out the invention]

[0009] The following describes in detail, with reference to the drawings, a solar power output prediction device, a power grid control system, a supply and demand control system, a solar radiation intensity prediction device, a learning device, a solar power output prediction method, and a solar power output prediction program according to the embodiment.

[0010] Figure 1 shows an example of the configuration of a power grid control system including a solar power output prediction device according to an embodiment. The solar power output prediction device 10 of this embodiment predicts solar radiation intensity with a desired time resolution using predicted values ​​of solar radiation intensity (or solar power output) acquired from an external source, and predicts solar power output using the predicted solar radiation intensity. Specifically, the solar power output prediction device 10 of this embodiment predicts a statistical quantity of solar radiation intensity with a desired time resolution using predicted values ​​of solar radiation intensity (or solar power output) acquired from an external source. Below, an example of predicting solar radiation intensity with a desired time resolution using predicted values ​​of solar radiation intensity will be given and explained, but as will be described later, solar radiation intensity may also be predicted using solar power output. The statistical quantity of solar radiation intensity predicted by the solar power output prediction device 10 is, for example, a percentile (in particular, the minimum value which is the 0th percentile, or the maximum value which is the 100th percentile, etc.) and an average value.

[0011] As shown in Figure 1, for example, the solar power output forecasting device 10, together with a power grid control device 20 that monitors and controls the voltage, current, etc. of the power grid, constitutes a power grid control system 30. The solar power output forecasting device 10 is not limited to the power grid control system 30, but may also be used for power supply and demand control, as will be described later, and the applications of the solar power output forecasting device 10 are not limited to this example. Here, we will first describe an example in which the solar power output predicted by the solar power output forecasting device 10 is used for monitoring and controlling the power grid, as shown in Figure 1.

[0012] The solar power output prediction device 10 includes a data acquisition unit 11, a data storage unit 12, a model generation unit 13, a trained model storage unit 14, a prediction unit 15, a transmission unit 16, and a display unit 17.

[0013] The data acquisition unit 11 acquires solar radiation intensity prediction values, which are predicted values ​​of solar radiation intensity, from the external prediction value provision system 40 and stores the acquired solar radiation intensity prediction values ​​in the data storage unit 12. The solar radiation intensity prediction values ​​provided by the external prediction value provision system 40 are information that associates the date and time with the prediction value. Hereinafter, the solar radiation intensity prediction values ​​acquired from the external prediction value provision system 40 will also be called external prediction values. External prediction values ​​may also be predicted values ​​of solar power generation output. The external prediction value provision system 40 may be a system that provides GSM (Global Spectral Model) data, a system that provides MSM (Meso-Scale Model) data, or any other system that provides solar radiation intensity prediction values. In addition, although one external prediction value provision system 40 is shown in Figure 1, there may be multiple external prediction value provision systems 40, and each of the multiple external prediction value provision systems 40 may provide solar radiation intensity prediction values ​​using different methods.

[0014] In addition, the data acquisition unit 11 acquires the actual value of solar radiation intensity associated with the date and time, which is the actual value obtained from the outside, from the performance value providing device 41, and stores the acquired actual value in the data storage unit 12. In this way, the data acquisition unit 11 acquires the predicted value of solar radiation intensity and the actual value of solar radiation intensity. As will be described later, the actual value of solar radiation intensity is used as correct answer data in the generation of the learned model for predicting solar radiation intensity. Hereinafter, the actual value of solar radiation intensity acquired from the performance value providing device 41 is also referred to as an external actual value. The actual value of solar radiation intensity may be a value measured by a solarimeter, or may be a value estimated from satellite images, Amedas measurement values, or the like. Further, as the external actual value, the actual value of solar power generation output may be used. The actual value of solar power generation output may be a measured value of solar power generation output, or may be an estimated value of solar power generation output estimated by a highly accurate method. The measured value of solar power generation output may be, for example, a value measured by a smart meter capable of separately measuring the generated power and the consumed power among smart meters that perform automatic meter reading of power consumption, or may be a value measured by a PCS (Power Conditioning System) in a solar power generation facility, or may be a value measured by other measuring devices for solar power generation output in a solar power generation facility. Examples of highly accurate methods for estimating solar power generation output include, but are not limited to, an estimation method using the measured value of solar power generation output or the actual value of solar radiation intensity at a location geographically close. In the following, an example in which the external actual value is the actual value of solar radiation intensity will be described.

[0015] The model generation unit 13 generates a learned model for predicting solar radiation intensity, specifically, a learned model for predicting the statistical quantity of solar radiation intensity, using the external predicted value and the actual value of solar radiation intensity stored in the data storage unit 12, and stores the generated learned model in the learned model storage unit 14. Details of the learned model will be described later.

[0016] The prediction unit 15 reads out the learned model stored in the learned model storage unit 14, and inputs the data corresponding to the prediction target period among the external prediction values stored in the data storage unit 12 into the learned model, thereby predicting the statistical amount of solar radiation intensity for the prediction target period. The prediction unit 15 predicts the statistical amount of photovoltaic power generation output using the prediction result of the statistical amount of solar radiation intensity, and outputs the prediction result of the statistical amount of photovoltaic power generation output to the transmission unit 16 and the display unit 17.

[0017] The transmission unit 16 transmits the prediction result of the statistical amount of photovoltaic power generation output to the power grid control device 20. The display unit 17 displays the prediction result of the statistical amount of photovoltaic power generation output. Further, the display unit 17 can also display the data used for the input of the prediction in the prediction unit 15.

[0018] Next, the operation of the photovoltaic power generation output prediction device 10 of the present embodiment will be described. The photovoltaic power generation output prediction device 10 of the present embodiment predicts the photovoltaic power generation output using external prediction values that are prediction values of solar radiation intensity, such as GSM data and MSM data. These external prediction values have a defined time resolution, for example, every hour. On the other hand, in the operation of the power grid, such as power grid monitoring and control, it is desired to predict the photovoltaic power generation output with a higher time resolution than the time resolution of these external prediction values, for example, every 30 minutes. The operation of the power grid includes power grid control for monitoring and controlling the power grid, and power supply and demand control for managing the balance between power demand (power consumption) and supply (generation amount).

[0019] The solar power output forecasting device 10 of this embodiment acquires actual solar radiation intensity values ​​with a higher temporal resolution than externally predicted values, and calculates statistical data of the actual values ​​for each time window of a predetermined length corresponding to the required temporal resolution. The solar power output forecasting device 10 then uses the time-series data of externally predicted values ​​for the time period from sunrise to sunset, and the "month" among the dates and times corresponding to the said time-series data as features, and uses the statistical data of the actual values ​​at the predicted date and time of the externally predicted values ​​as ground truth data to generate a trained model for each time window using supervised machine learning. The sunrise and sunset times may be fixed values ​​(for example, 6:00 and 18:00), monthly average values, or seasonal average values. By using this trained model, the solar power output forecasting device 10 predicts the statistical data of solar radiation intensity for each time window using the externally predicted values ​​for the date and time to be predicted. Since the length of the time window corresponds to the required temporal resolution, the solar power output forecasting device 10 of this embodiment can predict the statistical data of solar radiation intensity at the required temporal resolution, without depending on the temporal resolution of the externally predicted values. As a result, the solar power output forecasting device 10 can predict statistics of solar power output with the required time resolution. In power grid operation, it is desirable to predict solar power output under specific conditions, such as the most severe conditions. Therefore, by predicting percentiles of solar power output for each time period, such as the maximum value (100th percentile), minimum value (0th percentile), and median (50th percentile), it is possible to reflect this in the operation of the power grid. Accordingly, the statistics of solar radiation intensity are, for example, at least one of the maximum, minimum, and median values ​​of actual solar radiation intensity within a time period.

[0020] Figure 2 is a flowchart showing an example of the procedure for generating a trained model in the solar power output prediction device 10 of this embodiment. The solar power output prediction device 10 generates a trained model for each location to be predicted by, for example, the process exemplified in Figure 2. In Figure 2, an example of calculating the maximum and minimum values ​​as an example of solar radiation intensity statistics is shown, but the solar radiation intensity statistics are not limited to these, and may be either the maximum or minimum value, or the median, 25th percentile, 75th percentile, etc., or the mean value. Furthermore, there may be three or more values, such as the maximum, minimum, and median.

[0021] As shown in Figure 2, the solar power output prediction device 10 acquires predicted solar radiation intensity values ​​(external predicted values) from sunrise to sunset at multiple locations around the prediction point (step S1). Specifically, the data acquisition unit 11 periodically acquires external predicted values ​​from the external predicted value provision system 40 and stores them in the data storage unit 12, and the model generation unit 13 acquires external predicted values ​​by extracting and reading the external predicted values ​​from sunrise to sunset at multiple locations including the prediction point from the data acquisition unit 11. Alternatively, after the start of the process in Figure 2, the data acquisition unit 11 may acquire predicted solar radiation intensity values ​​from sunrise to sunset at multiple locations around the prediction point and store them in the data storage unit 12, and the model generation unit 13 may read this data from the data storage unit 12. The data acquired in step S1 is used as features in the generation of the trained model, as will be described later. Furthermore, sunrise and sunset times may be set in the solar power output prediction device 10 by being input by an operator or the like, or they may be transmitted from another device (not shown) and set in the solar power output prediction device 10. Similarly, prediction points and multiple points may be specified by being input by an operator or the like to the solar power output prediction device 10, or they may be specified from another device (not shown).

[0022] Here, we will explain the external prediction values ​​acquired by the solar power output prediction device 10 in step S1. External prediction values, including GSM data and MSM data, are provided with a spatial resolution of, for example, X km × X km (where X is a positive real number). That is, the external prediction values ​​are provided for each section, with X km × X km representing one section. Note that the spatial resolution of the external prediction values ​​is not limited to X km × X km; each section may be defined by latitude and longitude, and the size of each section does not have to be equal.

[0023] Figure 3 shows an example of the divisions of externally predicted values ​​in this embodiment. In Figure 3, 35 divisions of Xkm × Xkm are shown. If we refer to the identification information that identifies each division as the location number, then in Figure 3, 35 divisions from location number 1 to location number 35 are shown. Hereafter, the location of each division will also be referred to as a location. In this embodiment, when the division with location number 18 is designated as prediction location 200, which is the location to be predicted for solar power generation output, for example, a trained model is generated using time-series data of externally predicted values ​​from sunrise to sunset corresponding to prediction location 200, as well as the 34 divisions shown in Figure 3. By using externally predicted values ​​from multiple locations in this way, it is possible to generate a trained model that reflects not only the influence of prediction location 200 itself but also the influence of surrounding locations (divisions), thereby improving the accuracy of the trained model. When predicting the statistics of solar radiation intensity at prediction location 200, similarly, time-series data of externally predicted values ​​from sunrise to sunset corresponding to the 34 locations shown in Figure 3, as well as prediction location 200, are input into the trained model.

[0024] Figure 4 shows an example of external prediction values ​​used to generate a trained model. Figure 4 shows an example for a single day in June. As shown in Figure 4, the "month" corresponding to the external prediction values ​​and time-series data of predicted solar radiation intensity for each section, i.e., each point, from sunrise to sunset are used as features to generate the trained model. In the examples shown in Figures 3 and 4, external prediction values ​​for 35 locations are used as features, but Figures 3 and 4 are just examples, and the number of locations for which prediction values ​​are used as features can be one or more. In addition, the locations for which prediction values ​​are used as features are locations surrounding prediction point 200, but prediction point 200 itself may or may not be included. For example, only the external prediction values ​​for prediction point 200 may be used as features, or only the external prediction values ​​of multiple locations adjacent to prediction point 200 may be used as features. In the examples shown in Figures 3 and 4, external prediction values ​​are used for prediction point 200, two points adjacent to prediction point 200 in the longitude direction on each side, and three points adjacent to prediction point 200 in the latitudinal direction on each side. In this way, the number of points for which external prediction values ​​are used may differ between the latitudinal and longitude directions.

[0025] In the example shown in Figure 4, the external prediction values ​​are hourly predictions, but the time resolution of the external prediction values ​​is not limited to the example shown in Figure 4. Also, the sunrise and sunset times may be fixed regardless of the month, or they may be set for each month. In the example shown in Figure 4, the values ​​for June are used as an example, with sunrise at 6:00 and sunset at 18:00. Note that the sunrise and sunset times are not limited to this example. In the external prediction values, for example, the solar radiation intensity prediction value at 6:00 is the predicted average value from 6:00 to 7:00. Thus, the solar radiation intensity prediction values ​​for each time are predictions for the period from that time to the next time, so in Figure 4, the time series data from 6:00 to 17:00 is shown when the time period from sunrise to sunset is defined as 6:00 to 18:00.

[0026] Figure 5 is a schematic diagram showing an example of time-series data of external forecast values ​​as a graph. In Figure 5, time-series data of external forecast values ​​for a single location is shown. In Figure 5, similar to the example shown in Figure 4, the time period from sunrise to sunset is defined as 6:00 to 18:00, and 12 data points from 6:00 to 17:00 are shown. In step S1 shown in Figure 2, for example, if external forecast values ​​for 35 locations are used as shown in Figure 3, then for a given day's external forecast value, time-series data of 12 points as exemplified in this embodiment in Figure 5 will be acquired for 35 locations. In step S1, it is sufficient to acquire at least one day's worth of external forecasts for each "month".

[0027] Returning to the explanation of Figure 2, after step S1, the solar power output prediction device 10 sets the initial time (step S2). Specifically, the model generation unit 13 sets the time t to the initial time T1. T1 is the sunrise time mentioned above.

[0028] Next, the solar power output prediction device 10 obtains the maximum and minimum values ​​of the actual solar radiation intensity within a U-minute window, which is a time window at time t (step S3). The actual solar radiation intensity values ​​are actual solar radiation intensity values ​​obtained from the actual value providing device 41, and are measured or estimated values ​​obtained with a higher time resolution than the external prediction values. The actual solar radiation intensity values ​​are obtained by the data acquisition unit 11 and stored in the data storage unit 12. The model generation unit 13 reads the actual solar radiation intensity values ​​stored in the data storage unit 12 and obtains the maximum and minimum values ​​of the actual solar radiation intensity within a time window of U minutes starting from time t, thereby obtaining the maximum and minimum values ​​at time t.

[0029] Figure 6 shows an example of actual solar radiation intensity values ​​and the maximum and minimum values ​​within a time window. Figure 6 shows a magnified portion of the time period from sunrise to sunset. In the example shown in Figure 6, the time window U is set to 30, and the maximum and minimum actual values ​​within a 30-minute time window are calculated. In Figure 6, black circles indicate actual values, white squares indicate calculated maximum values, and white triangles indicate calculated minimum values. For example, the actual value 301, which is the maximum actual value during the 30 minutes from 9:00 to 9:30, is calculated as the maximum value 303 corresponding to 9:00, and the actual value 302, which is the minimum actual value during the 30 minutes from 9:00 to 9:30, is calculated as the minimum value 304 corresponding to 9:00. In this way, the maximum and minimum actual solar radiation intensity values ​​within the U-minute window at time t are calculated.

[0030] Returning to the explanation of Figure 2, after step S3, the solar power output forecasting device 10 generates a trained model using machine learning, with the external forecast value and "month" as features and the actual value as ground truth data (step S4). In detail, the model generation unit 13 generates a trained model using supervised learning, with the external forecast value obtained in step S1 and the "month" corresponding to that external forecast value as features, and the maximum and minimum values ​​of time t obtained in step S3 as ground truth data. The ground truth data used is the maximum and minimum values ​​of time t in the time period targeted by the trained model. This trained model is a trained model for inferring the maximum and minimum values, which are examples of statistics of solar radiation intensity at time t, from the external forecast value and the "month" corresponding to that external forecast value.

[0031] For example, if both externally predicted values ​​and actual solar radiation intensity values ​​are available for a given day in a given month, these can be used as a set of training data with features and ground truth data. A trained model can be generated by using multiple sets of these datasets. It is also possible that the time resolution of the externally predicted values ​​may be lower than the time resolution of the time window U. In this case, the same externally predicted values, which are features of the training data, will be used. For example, the time window U may be 30 minutes and the time resolution of the externally predicted values ​​may be 1 hour. In this case, for example, the average value of the externally predicted values ​​from 9:00 to 10:00 will be the predicted value at 9:00. For the training data from 9:00 to 9:30, the externally predicted value at 9:00 and the month will be used as features, and the maximum and minimum values ​​from 9:00 to 9:30 will be used as ground truth data. For the training data from 9:30 to 10:00, the externally predicted value at 9:00 and the month will be used as features, and the maximum and minimum values ​​from 9:30 to 10:00 will be used as ground truth data. In the above example, we explained a case where only external prediction values ​​for the time period targeted by the trained model were used as features for the training data. However, it is also possible to use all external prediction values ​​for the entire target day, or to use only a portion of them, as features for the training data.

[0032] In this way, the model generation unit 13 calculates statistical data of actual solar radiation intensity within a time period shorter than the time interval of the externally predicted solar radiation intensity forecast value, and uses the first externally predicted solar radiation intensity forecast value and the month corresponding to the first solar radiation intensity forecast value as input data, and uses training data including the input data and the above-mentioned statistical data of the month corresponding to the input data to generate a trained model for predicting statistical data of solar radiation intensity in each time period from the solar radiation intensity forecast value and the month using machine learning.

[0033] Any regression-type machine learning algorithm can be used as the supervised learning algorithm. For example, neural networks (including deep learning), decision trees, multiple regression, random forests, gradient boosting, and support vector regression can be used. A neural network consists of an input layer with multiple neurons, an intermediate layer (hidden layer) with multiple neurons, and an output layer with multiple neurons. The intermediate layer may be one or more.

[0034] Figure 7 is a schematic diagram illustrating an example of a neural network. For example, in a three-layer neural network like the one shown in Figure 7, when multiple inputs are input to the input layer (X1-X3), these values ​​are multiplied by weights W1 (w11-w16) and input to the hidden layer (Y1-Y2), and the result is further multiplied by weights W2 (w21-w26) and output from the output layer (Z1-Z3). This output result varies depending on the values ​​of weights W1 and W2.

[0035] In this embodiment, the relationship between the features and the ground truth data is learned by adjusting weights W1 and W2 so that the output from the output layer approaches the ground truth data when the features of the training data described above are input to the input layer. Note that, as mentioned above, the machine learning algorithm is not limited to a neural network.

[0036] Returning to the explanation of Figure 2, after step S4, the solar power output prediction device 10 advances the time by U minutes (step S5). Specifically, the model generation unit 13 updates the value of time t to t+U. Next, the solar power output prediction device 10 determines whether it has calculated up to the end time (step S6). Specifically, the model generation unit 13 determines whether time t has exceeded the end time T2. T2 is the sunset time mentioned above.

[0037] If the calculation is completed up to the end time (Step S6 Yes), the solar power output prediction device 10 terminates the process of generating the trained model. If the calculation is not completed up to the end time (Step S6 No), the solar power output prediction device 10 repeats the process from Step S3. Through the above process, a trained model corresponding to each time U minute from sunrise to sunset is generated. In other words, a trained model is generated for each time period. For example, if sunrise is 6:00, sunset is 18:00, and U is 30, a total of 24 trained models will be generated from 6:00 to 17:30, at 30-minute intervals. Note that U is not limited to 30 and should be determined according to the required time resolution.

[0038] If multiple prediction locations are anticipated, a trained model for each prediction location is created by performing the process shown in Figure 2 for each location.

[0039] Next, we will explain how to predict solar radiation intensity statistics using a trained model. Figure 8 is a flowchart showing an example of the procedure for predicting solar radiation intensity statistics in this embodiment. In Figure 8, we will explain an example in which the maximum and minimum values ​​are predicted as statistics, similar to the example shown in Figure 2.

[0040] The solar power output forecasting device 10 acquires predicted solar radiation intensity values ​​(external predicted values) from sunrise to sunset at multiple locations around the forecasting site (step S11). Specifically, the forecasting unit 15 acquires the external predicted values ​​by extracting and reading the predicted solar radiation intensity values ​​(external predicted values) from sunrise to sunset at multiple locations around the forecasting site on the forecasting day from the external predicted values ​​stored in the data storage unit 12. The forecasting day, forecasting site, and multiple locations may be specified by input to the solar power output forecasting device 10 by an operator or the like, or they may be specified by another device not shown.

[0041] Next, the solar power output prediction device 10 sets the initial time (step S12). Specifically, the prediction unit 15 sets time t to the initial time T1. Next, the solar power output prediction device 10 retrieves the trained model at time t (step S13). Specifically, the prediction unit 15 retrieves the trained model by reading the trained model corresponding to time t from the trained model storage unit 14.

[0042] Next, the solar power output prediction device 10 inputs the external prediction values ​​and the "month" into the trained model and predicts the maximum and minimum values ​​of solar radiation intensity within the U-minute window at time t (step S14). Specifically, the prediction unit 15 inputs the external prediction values ​​acquired in step S11 into the trained model and obtains the maximum and minimum values ​​of solar radiation intensity output from the trained model, thereby predicting the maximum and minimum values ​​of solar radiation intensity at time t.

[0043] In this way, the prediction unit 15 inputs the second solar radiation intensity prediction value, which is the solar radiation intensity prediction value for the day to be predicted, and the month corresponding to the second solar radiation intensity prediction value, both acquired by the data acquisition unit 11, into the trained model, and predicts the solar radiation intensity statistics for each time period as the output of the trained model.

[0044] Next, the solar power output forecasting device 10 advances the time by U minutes (step S15). Specifically, the forecasting unit 15 updates the value of time t to t+U. Next, the solar power output forecasting device 10 determines whether it has calculated up to the end time (step S16). Specifically, the forecasting unit 15 determines whether time t has exceeded the end time T2.

[0045] If the calculation is completed up to the end time (step S16 Yes), the solar power output prediction device 10 terminates the prediction process. If the calculation is not completed up to the end time (step S16 No), the solar power output prediction device 10 repeats the process from step S13. Through the above process, the maximum and minimum values ​​of solar radiation intensity for each U-minute interval from sunrise to sunset at the prediction point on the prediction target day are calculated. The prediction unit 15 also uses the predicted results of the maximum and minimum values ​​of solar radiation intensity to calculate predicted values ​​of the maximum and minimum values ​​of solar power output, and outputs the calculated predicted values ​​to the transmission unit 16. Any general method can be used to calculate solar power output from solar radiation intensity, such as multiplying the solar radiation intensity by the rated capacity and coefficients of the solar power generation equipment. Alternatively, a conversion coefficient for converting solar radiation intensity to solar power output may be determined based on actual values, and this conversion coefficient may be used to calculate solar power output from solar radiation intensity. The transmission unit 16 transmits the predicted values ​​calculated by the prediction unit 15 to the power system control device 20. Alternatively, the transmitting unit 16 may transmit the statistical amount of solar radiation intensity predicted by the prediction unit 15 to the power system control device 20, and the power system control device 20 may obtain a predicted value of solar power generation output from the statistical amount of solar radiation intensity.

[0046] Furthermore, the prediction unit 15 passes the predicted maximum and minimum values ​​of solar radiation intensity and the maximum and minimum values ​​of solar power generation output to the display unit 17, which displays the predicted maximum and minimum values ​​of solar radiation intensity and the maximum and minimum values ​​of solar power generation output. In the example shown in Figure 8, the maximum and minimum values ​​of solar radiation intensity from sunrise to sunset were calculated. However, if you want to predict the maximum and minimum values ​​of solar radiation intensity at a specific time on a given day, instead of using sunrise and sunset times for T1 in step S12 and T2 in step S16, respectively, you can set the time to be predicted as T1 and T2 to a value between T1 and T1+U.

[0047] Figure 9 shows an example of the external predicted values ​​and the predicted maximum and minimum values ​​of solar radiation intensity for the day to be predicted in this embodiment. The upper part of Figure 9 shows the external predicted values ​​that serve as input data in the prediction processing of the prediction unit 15, i.e., the external predicted values ​​for the day to be predicted obtained in step S11, and the lower part of Figure 9 shows the predicted maximum and minimum values ​​of solar radiation intensity. In Figure 9, sunrise is set to 6:00, sunset to 18:00, and the time window U is set to 30. As can be seen from Figure 9, the time resolution of the external predicted values ​​is 1 hour, but the predicted maximum and minimum values ​​of solar radiation intensity, which are the prediction results, are obtained every 30 minutes. Thus, in this embodiment, the prediction results of solar radiation intensity can be obtained with a higher resolution than the time resolution of the external predicted values. For this reason, the solar power generation output can also be predicted with a higher resolution than the time resolution of the external predicted values.

[0048] The display unit 17 can display external predicted values ​​shown in the upper part of Figure 9, and predicted maximum and minimum solar radiation intensity values ​​shown in the lower part of Figure 9. Similarly, although not shown in the illustration, the display unit 17 can display predicted maximum and minimum solar power generation output values.

[0049] In the example above, externally predicted values ​​for the entire time period from sunrise to sunset were used as features. That is, in the example above, the first solar radiation intensity prediction value, which is an externally predicted value used to generate the trained model, and the second solar radiation intensity prediction value, which is an externally predicted value input to the trained model in the prediction process, are time series data from the first time to the second time of day, where the first time is sunrise and the second time is sunset. The first and second times that determine the length of the time series data of the externally predicted values ​​are not limited to these, and externally predicted values ​​for one or more time frames that include the time t to be predicted may be used. For example, to generate a trained model corresponding to 9 o'clock, only the value of the externally predicted value for 9 o'clock may be used as a feature, or the values ​​of the externally predicted values ​​for 8 o'clock, 9 o'clock and 10 o'clock may be used.

[0050] Alternatively, the solar power output prediction device 10 may obtain multiple types of predicted solar radiation intensity as external predicted values, create training data using these multiple types of predicted solar radiation intensity values ​​to create a trained model, and then input the multiple types of predicted solar radiation intensity values ​​for the target day into the created trained model to predict solar radiation intensity, i.e., the statistics of solar radiation intensity. In this case, even if the time resolution of the multiple types of predicted solar radiation intensity values ​​is different, the predicted values ​​from sunrise to sunset can be simply listed and input as features. Also, even if the spatial resolution of the multiple types of predicted solar radiation intensity values ​​is different, one or more locations to be used in the trained model corresponding to each prediction location can be determined for each type, and the time series data of one or more external predicted values ​​for each type can be input as features. By performing machine learning using these multiple types of predicted solar radiation intensity values, external predicted values ​​that are highly correlated with the solar radiation intensity at the prediction location at time t (effective for learning) will be automatically selected and used for prediction.

[0051] In the example above, the external prediction value was explained as the predicted value of solar radiation intensity, but as mentioned above, the external prediction value may also be the solar power generation output. Furthermore, in the example above, a trained model was generated to predict the statistics of solar radiation intensity using actual values ​​of solar radiation intensity as the ground truth data, but a trained model to predict the statistics of solar radiation intensity or the statistics of solar power generation output may also be generated using actual values ​​of solar power generation output as the ground truth data. In this case, the external measured values ​​may be measured values ​​from smart meters of the power generation output at the solar power generation facility, measured values ​​from PCS, etc., or estimated values ​​of solar power generation output estimated by a highly accurate method may be used.

[0052] The combination of solar radiation intensity and solar power output can be summarized as follows. First, (1) external predicted values ​​(input data for the trained model), (2) external actual values ​​(ground truth data), and (3) statistics of the target of prediction are classified as follows. (1) External predictions (1a) Solar radiation (1b) Solar power output (2) External performance values (2a) Solar radiation intensity (2b) Solar power output (3) Statistics to be predicted (3a) Solar radiation (3b) Solar power output

[0053] When obtaining (3a), one of two patterns is used: (1a)-(2a) or (1b)-(2a). When obtaining (3b), one of four patterns is used: (1a)-(2a), (1a)-(2b), (1b)-(2a), or (1b)-(2b). Note that when using either of the two patterns (1a)-(2a) or (1b)-(2a) to obtain (3b), it is necessary to convert the solar radiation statistic, which is the output of the trained model, into a statistic of the photovoltaic power output. It is also possible to use a mixture of (2a) and (2b). When using a mixture of (2a) and (2b), the solar radiation statistic for (2a) is converted into the photovoltaic power output.

[0054] Furthermore, in the above example, a trained model was generated to predict the statistics of solar radiation intensity using actual values ​​of solar radiation intensity as ground truth data. However, a trained model to predict the statistics of solar power generation output may also be generated using actual values ​​of solar power generation output as ground truth data. In this case, the model generation unit 13 can use solar power generation output calculated using measured values ​​from smart meters of the solar power generation facility as actual measured values. If measured values ​​of solar radiation intensity are available, these solar radiation intensity values ​​may be converted to solar power generation output and used as actual values ​​of solar power generation output. When a trained model to predict the statistics of solar power generation output is generated, the prediction unit 15 can obtain predicted values ​​of the statistics of solar power generation output by inputting external predicted values ​​for multiple locations corresponding to the prediction location on the prediction target day and the corresponding "month" into this trained model. In other words, the trained model in this embodiment only needs to be a model for predicting a value indicating solar power generation output, and the value indicating solar power generation output may be the solar power generation output itself or solar radiation intensity.

[0055] Here, the hardware configuration of the solar power output prediction device 10 will be described. In this embodiment, the solar power output prediction device 10 functions as a computer system when a solar power output prediction program, which is a program describing the processing in the solar power output prediction device 10, is executed on the computer system. Figure 10 is a diagram showing an example of the configuration of a computer system that realizes the solar power output prediction device 10 of this embodiment. As shown in Figure 10, this computer system comprises a control unit 101, an input unit 102, a storage unit 103, a display unit 104, a communication unit 105, and an output unit 106, which are connected via a system bus 107.

[0056] In Figure 10, the control unit 101 is a processor such as a CPU (Central Processing Unit) and executes a solar power output prediction program that describes the processing in the solar power output prediction device 10 of this embodiment. The input unit 102 consists of, for example, a keyboard and mouse and is used by the user of the computer system to input various information. The storage unit 103 includes various types of memory such as RAM (Random Access Memory) and ROM (Read Only Memory), and storage devices such as a hard disk and stores the program to be executed by the control unit 101, necessary data obtained during the processing, etc. The storage unit 103 is also used as a temporary storage area for the program. The display unit 104 consists of a display, LCD (Liquid Crystal Display Panel), etc., and displays various screens to the user of the computer system. The communication unit 105 is a receiver and transmitter that perform communication processing. The output unit 106 is a printer, etc. Note that Figure 10 is an example, and the configuration of the computer system is not limited to the example in Figure 10.

[0057] Here, an example of the operation of the computer system until the solar power output prediction program of this embodiment becomes executable will be described. In the computer system with the above configuration, for example, the solar power output prediction program is installed in the storage unit 103 from a CD-ROM or DVD-ROM set in a CD (Compact Disc)-ROM drive or DVD (Digital Versatile Disc)-ROM drive (not shown). When the solar power output prediction program is executed, the solar power output prediction program read from the storage unit 103 is stored in the storage unit 103. In this state, the control unit 101 performs processing as the solar power output prediction device 10 of this embodiment according to the program stored in the storage unit 103.

[0058] In the above description, a program describing the processing in the solar power output prediction device 10 is provided on a CD-ROM or DVD-ROM as the recording medium. However, the explanation is not limited to this, and depending on the configuration of the computer system, the capacity of the program to be provided, a program provided via a transmission medium such as the Internet via the communication unit 105 may be used, for example.

[0059] The model generation unit 13 and prediction unit 15 shown in Figure 1 are realized by the execution of a solar power generation output prediction program stored in the storage unit 103 shown in Figure 10 by the control unit 101 shown in Figure 10. The data storage unit 12 and trained model storage unit 14 shown in Figure 1 are part of the storage unit 103 shown in Figure 10. The data acquisition unit 11 and transmission unit 16 shown in Figure 1 are realized by the communication unit 105 and control unit 101 shown in Figure 10. The display unit 17 shown in Figure 1 is realized by the display unit 104 shown in Figure 10. The solar power generation output prediction device 10 may be realized by multiple computer systems. Alternatively, the solar power generation output prediction device 10 may be realized by a cloud system.

[0060] For example, the solar power output prediction program of this embodiment causes the computer system to perform the following steps: calculate statistical data of actual values ​​indicating solar power output within a time period shorter than the time interval of the predicted solar radiation intensity values; and use a first predicted solar radiation intensity value and the month corresponding to the first predicted solar radiation intensity value as input data, and use training data including the input data and the statistical data of the month corresponding to the input data to generate a trained model for predicting statistical data of values ​​indicating solar power output in a given time period from the predicted solar radiation intensity value and the month using machine learning. Furthermore, the solar power output prediction program causes the computer system to perform the following steps: input a second predicted solar radiation intensity value, which is the predicted solar radiation intensity value for the day to be predicted, and the month corresponding to the second predicted solar radiation intensity value into the trained model, thereby predicting statistical data of values ​​indicating solar power output for each time period as the output of the trained model.

[0061] <Example 1> In the example shown in Figure 1, the solar power output prediction device 10 performed both the generation of a trained model and the prediction using the trained model. However, the generation of the trained model and the prediction using the trained model could be performed by separate devices.

[0062] Figure 11 shows another example configuration of the power grid control system of this embodiment. In the configuration example shown in Figure 11, the generation of the trained model and the prediction using the trained model are performed by separate devices. Components having the same functions as the solar power output prediction device 10 shown in Figure 1 are denoted by the same reference numerals as in Figure 1, and redundant explanations are omitted. In the configuration example shown in Figure 11, the generation of the trained model is performed by the learning device 70, and the prediction using the trained model is performed by the solar power output prediction device 10a. The power grid control system 30a shown in Figure 11 comprises the solar power output prediction device 10a and the power grid control device 20. The learning device 70 comprises a data acquisition unit 71, a data storage unit 72, a model generation unit 73, and a trained model storage unit 74. The functions of the data acquisition unit 71, data storage unit 72, model generation unit 73, and trained model storage unit 74 are the same as the functions of the data acquisition unit 11, data storage unit 12, model generation unit 13, and trained model storage unit 14 related to the generation of the trained model in the solar power output prediction device 10 described above.

[0063] The solar power output forecasting device 10a is the same as the solar power output forecasting device 10 but with the model generation unit 13 and data storage unit 12 removed, and performs forecasting processing in the same way as the solar power output forecasting device 10. The trained model storage unit 14 stores the trained model generated by the learning device 70, that is, the trained model stored in the trained model storage unit 74 of the learning device 70. In the solar power output forecasting device 10a, the external forecast values ​​for the forecast target day are input to the forecasting unit 15 from the data acquisition unit 11. However, the solar power output forecasting device 10a may also include a data storage unit (not shown) in which the data acquisition unit 11 stores the acquired external forecast values ​​in the data storage unit, and the forecasting unit 15 reads the external forecast values ​​corresponding to the forecast target day from the data storage unit.

[0064] In Figure 11, the learning device 70 is shown as being outside the power grid control system 30a, but the learning device 70 may also be included within the power grid control system 30a. The learning device 70 and the solar power output prediction device 10a are also implemented by a computer system, similar to the solar power output prediction device 10.

[0065] <Modification 2> Figure 1 illustrates the solar power output forecasting device 10, which, together with the power grid control device 20, constitutes the power grid control system 30, as an example. However, the above-described configuration and operation can also be applied to, for example, an Energy Management System (EMS) that monitors and controls the balance of power supply and demand. Figure 12 shows an example configuration of a supply and demand control system including the solar power output forecasting device of this embodiment. The supply and demand control system 60 shown in Figure 12 comprises a solar power output forecasting device 10b and a supply and demand control device 50. The solar power output forecasting device 10b is the same as the solar power output forecasting device 10 shown in Figure 1, except that an aggregation unit 18 is added and a display unit 17a is provided instead of a display unit 17. Components having the same functions as the solar power output forecasting device 10 shown in Figure 1 are denoted by the same reference numerals as in Figure 1, and redundant explanations are omitted.

[0066] In the supply and demand control device 50, it is generally desirable to predict solar power output over a wider area than that of the power grid monitoring and control. For example, it is desirable to predict solar power output for the entire area under the jurisdiction of a power company or for units such as prefectures, which are several divisions of the area under its jurisdiction. For this reason, in the solar power output prediction device 10b shown in Figure 12, the aggregation unit 18 aggregates the prediction results from multiple prediction points to calculate the prediction result of solar power output over a wide area corresponding to multiple prediction points, and transmits the prediction result to the supply and demand control device 50. The aggregation unit 18 outputs the calculated prediction result to the display unit 17a. The display unit 17a, like the display unit 17 shown in Figure 1, can display the prediction results for each prediction point output from the prediction unit 15, and can also display the prediction result of solar power output over a wide area aggregated by the aggregation unit 18. The solar power output prediction device 10b is also implemented by a computer system, similar to the solar power output prediction device 10. Alternatively, the solar power output prediction device 10 shown in Figure 1 may transmit predicted values ​​of the solar power output statistics to the supply and demand control device 50, and the supply and demand control device 50 may aggregate the predicted values. Alternatively, the transmission unit 16 may transmit the predicted solar radiation intensity statistics predicted by the prediction unit 15 to the supply and demand control device 50, and the supply and demand control device 50 may obtain predicted values ​​of the solar power output from the solar radiation intensity statistics. Furthermore, similar to the first modification, a learning device and a solar power output prediction device that predicts the solar power output may be provided separately.

[0067] Furthermore, the prediction results from the above-described solar power output prediction devices 10, 10a, and 10b can also be used for predicting market prices in electricity trading, planning market transactions, and more. In addition, although the above describes an example of predicting solar radiation intensity for the purpose of predicting solar power output, it is also possible to predict solar radiation intensity, i.e., predict the statistical amount of solar radiation intensity, without predicting the solar power output. In this case, the solar power output prediction device 10 shown in Figure 1 functions as a solar radiation intensity prediction device. In this case, the prediction unit 15 only needs to predict the statistical amount of solar radiation intensity, and it is not necessary to calculate solar power output from the solar radiation intensity. The statistical amount of solar radiation intensity predicted by the solar radiation intensity prediction device may be used in other devices to predict solar power output, for planning the installation of solar power generation facilities, or for other purposes such as agriculture.

[0068] As described above, the solar power output prediction devices 10, 10b and the learning device 70 of this embodiment acquire feature quantities including external prediction values, which are predicted values ​​of solar radiation intensity for one or more locations acquired from the outside, and the "month" corresponding to the external prediction values, and statistics of actual values ​​indicating solar power output within a time period shorter than the time interval of the external prediction values, as training data. Then, the solar power output prediction devices 10, 10b and the learning device 70 use the training data to generate a trained model for predicting statistics of solar radiation intensity from the external prediction values ​​and the corresponding "month" using machine learning. The solar power output prediction devices 10, 10a and 10b of this embodiment input the external prediction values ​​and the "month" corresponding to the external prediction values ​​into the trained model to predict statistics of values ​​indicating solar power output with a time resolution corresponding to the time frame. This makes it possible to predict solar power output with a time resolution higher than the time resolution of the acquired predicted values ​​of solar radiation intensity.

[0069] The configurations shown in the above embodiments are examples only, and it is possible to combine them with other known technologies, combine different embodiments, and omit or modify parts of the configuration without departing from the gist of the invention. [Explanation of Symbols]

[0070] 10,10a,10b Solar power output prediction device, 11,71 Data acquisition unit, 12,72 Data storage unit, 13,73 Model generation unit, 14,74 Trained model storage unit, 15 Prediction unit, 16 Transmission unit, 17,17a Display unit, 18 Aggregation unit, 20 Power grid control device, 30,30a Power grid control system, 40 External prediction value provision system, 41 Actual value provision device, 50 Supply and demand control device, 60 Supply and demand control system, 70 Learning device.

Claims

1. A data acquisition unit that acquires solar radiation intensity prediction values, which are one or more data points corresponding to one or more times within the period from sunrise to sunset among the predicted solar radiation intensity values ​​obtained as time series data, and actual values ​​indicating solar power generation output for time periods shorter than the time interval of the time series data, A model generation unit calculates statistical quantities of the actual values ​​within each of the aforementioned time periods, takes a first solar radiation intensity prediction value, which is the solar radiation intensity prediction value, and the month corresponding to the first solar radiation intensity prediction value as input data, and uses training data including the input data and the statistical quantities corresponding to the input data to generate a trained model for predicting statistical quantities of values ​​indicating solar power generation output in the aforementioned time period from the solar radiation intensity prediction value and the month by machine learning. A prediction unit predicts a statistical value representing the solar power generation output for each time period by inputting a second solar radiation intensity prediction value, which is the solar radiation intensity prediction value for the day to be predicted, and the month corresponding to the second solar radiation intensity prediction value, obtained by the data acquisition unit, into the trained model, and as the output of the trained model, Equipped with, The aforementioned actual values ​​include one or more data points corresponding to one or more times within the aforementioned time period, A solar power generation output prediction device characterized in that the time corresponding to the solar radiation intensity prediction value used by the model generation unit and the prediction unit includes a time within the corresponding time period.

2. The solar power output prediction device according to claim 1, characterized in that the aforementioned statistical quantity is at least one of percentiles and mean values.

3. The solar power output forecasting device according to claim 2, characterized in that the statistic is at least one of the maximum, minimum, and median values ​​of the actual values ​​within the time period.

4. The solar power output forecasting device according to claim 2, characterized in that the statistical quantities are the maximum and minimum values ​​of the actual values ​​within the time period.

5. The solar power output forecasting device according to any one of claims 1 to 4, characterized in that the first solar radiation intensity forecast value and the second solar radiation intensity forecast value are time-series data from a first time to a second time on a day.

6. The solar power output prediction device according to claim 5, characterized in that the first time is the time of sunrise and the second time is the time of sunset.

7. The solar power output prediction device according to any one of claims 1 to 6, characterized in that the first solar radiation intensity prediction value and the second solar radiation intensity prediction value are solar radiation intensity prediction values ​​for one or more locations.

8. The trained model is generated for each prediction point, which is the point for which the statistical value representing the solar power generation output is to be predicted. The solar power output forecasting device according to claim 7, characterized in that the one or more locations are a plurality of locations including the forecasting location.

9. The solar power output prediction device according to any one of claims 1 to 8, characterized in that the trained model is generated for each time period.

10. The value indicating the output of solar power generation is solar radiation intensity. The solar power output prediction device according to any one of claims 1 to 9, characterized in that the prediction unit calculates a statistical amount of solar power output using the predicted statistical amount for each time period.

11. The solar power output prediction device according to any one of claims 1 to 9, characterized in that the value indicating the solar power output is the solar power output itself.

12. A data acquisition unit that acquires solar radiation intensity prediction values, which are one or more data points corresponding to one or more times within the period from sunrise to sunset among the predicted solar radiation intensity values ​​obtained as time series data, and actual values ​​indicating solar power generation output for time periods shorter than the time interval of the time series data, A trained model storage unit stores a trained model for predicting statistical quantities within a time period shorter than the time interval of the time series data representing solar power generation output from predicted solar radiation intensity and the moon for each of the aforementioned time periods. A prediction unit that inputs the predicted solar radiation intensity value and month for the day to be predicted, obtained by the data acquisition unit, into the trained model, and predicts a statistical value representing the solar power generation output for each time period as the output of the trained model. Equipped with, The aforementioned actual values ​​include one or more data points corresponding to one or more times within the aforementioned time period, A solar power generation output prediction device characterized in that the time corresponding to the solar radiation intensity prediction value used by the prediction unit includes a time within the corresponding time period.

13. A solar power output prediction device according to any one of claims 1 to 12, A power system control device that monitors and controls the power system using statistical values ​​of the predicted solar power output predicted by the solar power output prediction device, A power grid control system characterized by comprising the following features.

14. A solar power output prediction device according to any one of claims 1 to 12, A power supply and demand control device that controls the supply and demand of electricity using a statistical quantity of a value indicating the solar power output predicted by the solar power output prediction device, A supply and demand control system characterized by comprising the following features.

15. A data acquisition unit that acquires solar radiation intensity prediction values, which are one or more data points corresponding to one or more times within the period from sunrise to sunset among the predicted solar radiation intensity values ​​obtained as time series data, and actual values ​​indicating solar power generation output for time periods shorter than the time interval of the time series data, A model generation unit calculates statistical quantities of the actual values ​​within each of the aforementioned time periods, takes a first solar radiation intensity prediction value, which is the solar radiation intensity prediction value, and the month corresponding to the first solar radiation intensity prediction value as input data, and uses training data including the input data and the statistical quantities corresponding to the input data to generate a trained model for predicting statistical quantities of values ​​indicating solar power generation output in the aforementioned time period from the solar radiation intensity prediction value and the month by machine learning. Equipped with, The aforementioned actual values ​​include one or more data points corresponding to one or more times within the aforementioned time period, A learning device characterized in that the time corresponding to the solar radiation intensity prediction value used by the model generation unit includes a time within the corresponding time period.

16. A method for predicting solar power output in a solar power output prediction device, A first acquisition step involves obtaining solar radiation intensity prediction values, which are one or more data points corresponding to one or more times within the period from sunrise to sunset, from the predicted solar radiation intensity values ​​obtained as time series data. A second acquisition step involves obtaining actual values ​​of the solar power generation output for each time period shorter than the time interval of the aforementioned time series data, A calculation step for each of the aforementioned time periods, which involves calculating a statistical amount of the actual value within the aforementioned time period, For each of the aforementioned time periods, a generation step is to generate a trained model for predicting a statistical value representing solar power output for the aforementioned time period from the solar radiation intensity prediction value and the month, using machine learning with input data including a first solar radiation intensity prediction value and the month corresponding to the first solar radiation intensity prediction value, and training data including the input data and the statistical value corresponding to the input data. A prediction step in which a statistical quantity of a value indicating solar power generation output is predicted for each time period by inputting a second solar radiation intensity prediction value, which is the solar radiation intensity prediction value for the day to be predicted, and the month corresponding to the second solar radiation intensity prediction value, obtained by the first acquisition step, into the trained model, and Includes, The aforementioned actual values ​​include one or more data points corresponding to one or more times within the aforementioned time period, A method for predicting solar power output, characterized in that the time corresponding to the predicted solar radiation value used in the generation step and the prediction step includes a time within the corresponding time period.

17. In the computer system, A first acquisition step involves obtaining solar radiation intensity prediction values, which are one or more data points corresponding to one or more times within the period from sunrise to sunset, from the predicted solar radiation intensity values ​​obtained as time series data. A second acquisition step involves obtaining actual values ​​of the solar power generation output for each time period shorter than the time interval of the aforementioned time series data, A calculation step for each of the aforementioned time periods, which involves calculating a statistical amount of the actual value within the aforementioned time period, For each of the aforementioned time periods, a generation step is to generate a trained model for predicting a statistical value representing solar power output for the aforementioned time period from the solar radiation intensity prediction value and the month, using machine learning with input data including a first solar radiation intensity prediction value and the month corresponding to the first solar radiation intensity prediction value, and training data including the input data and the statistical value corresponding to the input data. A prediction step in which a statistical quantity of a value indicating solar power generation output is predicted for each time period by inputting a second solar radiation intensity prediction value, which is the solar radiation intensity prediction value for the day to be predicted, and the month corresponding to the second solar radiation intensity prediction value, obtained by the first acquisition step, into the trained model, and Make it run, The aforementioned actual values ​​include one or more data points corresponding to one or more times within the aforementioned time period, A solar power generation output prediction program characterized in that the time corresponding to the solar radiation intensity prediction value used in the generation step and the prediction step includes a time within the corresponding time period.

Images