Solar radiation intensity prediction device and solar radiation intensity prediction system

Abstract

This solar radiation intensity prediction device predicts solar radiation intensity over a long period of time, which is a period of at least one month, using weather data prediction values, actual solar radiation intensity values, and actual weather data values. The solar radiation intensity prediction device is used to perform long-term solar radiation intensity prediction so as to assist in creation of a long-term energy operation plan. The solar radiation intensity prediction device comprises: a learning model generation unit that uses actual solar radiation intensity values, which are correct answer data, and a feature amount to generate, by machine learning, a trained model for inferring a prediction value of solar radiation intensity over a long period of time; a prediction unit that acquires, upon input of the feature amount to the learned model, a solar radiation intensity prediction value by performing inference using the learned model; a solar radiation intensity statistic calculation unit that calculates, from the solar radiation intensity prediction value, a statistic of solar radiation intensity for compensating for a prediction error that occurs with long-term prediction; and a transmission unit that transmits the statistic.

Description

Solar radiation intensity forecasting device and solar radiation intensity forecasting system

[0001] The present disclosure relates to a solar radiation intensity prediction device, and more particularly to a solar radiation intensity prediction device that covers a long period of time.

[0002] Currently, renewable energy sources such as solar and wind power are attracting attention. However, the amount of power generated by solar power generation fluctuates significantly due to factors such as seasonal variations in the amount of solar radiation, the angle of incidence of sunlight, and temperature, as well as sudden changes in the weather. Therefore, in order to utilize the power generated by solar power generation more efficiently, a technology is needed that can accurately predict the future amount of power generated by solar power generation.

[0003] Patent Document 1 discloses a technology for accurately predicting future power generation amounts by responding not only to sudden changes in weather but also to periodic weather such as seasonal fluctuations and changes in the properties of solar power generation panels.

[0004] Furthermore, Patent Document 2 discloses a technology for improving the accuracy of predictions of the amount of power generated by photovoltaic power generation, the amount of solar radiation, and the wind speed through machine learning.

[0005] JP 2013-99143 A International Publication No. 2016 / 121202 JP 2013-44572 A

[0006] In order to create an energy operation plan for several months, taking into account renewable energy output and demand several months into the future, it is necessary to make long-term forecasts of renewable energy output and electricity demand, particularly solar power generation output.

[0007] The prediction techniques disclosed in Patent Documents 1 and 2 are intended for short-term predictions of a few hours or days in advance, and are unable to predict photovoltaic power generation output, i.e., solar radiation intensity, for long-term predictions of one month or more. Therefore, it is not possible to create long-term energy operation plans. Furthermore, it is considered difficult to accurately predict solar radiation intensity simply by applying the conventional techniques developed for short-term predictions to long-term predictions.

[0008] The present disclosure has been made to solve the above-mentioned problems, and aims to provide a solar radiation intensity prediction device that enables long-term solar radiation intensity prediction and can assist in creating long-term energy operation plans.

[0009] A solar radiation intensity prediction device according to the present disclosure is a solar radiation intensity prediction device that predicts solar radiation intensity over a long period of time, which is one month or more, using predicted values ​​of meteorological data, actual solar radiation intensity values, and actual meteorological data values, and the device includes: a learning model generation unit that generates a trained model by machine learning using the actual solar radiation intensity values, which are correct data, and feature values, to infer predicted values ​​of the solar radiation intensity over the long period; and a feature generation unit that generates calculated feature values ​​of the feature values ​​other than past values ​​of the actual meteorological data values ​​and the predicted values ​​of the meteorological data. the long-term prediction system includes a prediction unit that predicts the long-term solar radiation intensity to obtain a solar radiation intensity predicted value; a memory unit that stores the meteorological data actual value, the solar radiation intensity actual value, the solar radiation intensity predicted value, the feature values, and the trained model; a solar radiation intensity statistics calculation unit that calculates statistics of the solar radiation intensity from the solar radiation intensity predicted value to compensate for prediction errors that occur during long-term prediction; and a transmission unit that transmits the statistics, wherein the prediction unit inputs the feature values ​​into the trained model, thereby performing inference using the trained model to obtain the solar radiation intensity predicted value.

[0010] The solar radiation intensity prediction device according to the present disclosure makes it possible to predict solar radiation intensity over a long period of time, and can assist in creating a long-term energy operation plan.

[0011] 1 is a conceptual diagram of a solar radiation intensity prediction system according to a first embodiment. FIG. 2 is a block diagram illustrating the configuration of a solar radiation intensity prediction device according to the first embodiment. FIG. 3 is a flowchart illustrating processing during learning in the solar radiation intensity prediction device according to the first embodiment and processing by the learning device in the solar radiation intensity prediction device according to the second embodiment. FIG. 4 is a diagram illustrating theoretical solar radiation intensity. FIG. 5 is a diagram illustrating learning data including feature quantities and actual solar radiation intensity values. FIG. 6 is a flowchart illustrating processing during prediction in the solar radiation intensity prediction device according to the first embodiment and prediction processing by the prediction device in the solar radiation intensity prediction device according to the second embodiment. FIG. 7 is a flowchart illustrating processing for calculating statistics in the solar radiation intensity prediction device according to the first embodiment and processing for calculating statistics by the prediction device in the solar radiation intensity prediction device according to the second embodiment. FIG. 8 is a block diagram illustrating the configuration of a solar radiation intensity prediction device according to the second embodiment. FIG. 9 is a diagram illustrating a hardware configuration for realizing the solar radiation intensity prediction devices according to the first and second embodiments of the present disclosure. FIG. 10 is a diagram illustrating a hardware configuration for realizing the solar radiation intensity prediction devices according to the first and second embodiments of the present disclosure.

[0012] <Introduction> Renewable energy sources such as solar and wind power are attracting attention as a way to realize a decarbonized society. However, renewable energy output fluctuates greatly depending on the weather, and surplus electricity, i.e., electricity generated by renewable energy minus electricity demand, has traditionally been stored using storage batteries. However, storage batteries cannot store large amounts of electricity and are prone to self-discharge, making them unsuitable for long-term storage across seasons, spanning several months.

[0013] On the other hand, hydrogen and methane are considered suitable for long-term storage because, unlike storage batteries, they do not self-discharge. Therefore, in recent years, energy operation plans have been considered to achieve so-called peak shift, in which surplus electricity from renewable energy sources is used to generate hydrogen or methane, which is then stored in tanks for long periods and used for power generation during periods of power shortage.

[0014] The technology disclosed herein predicts the amount of solar power generation (PV amount) and solar radiation intensity output at a certain location over the long term, for example, three months from now, making it possible to predict whether surplus power or power shortages will occur several months from now, and to use this information in long-term energy operation planning.

[0015] <Embodiment 1> <Device Configuration> Fig. 1 is a conceptual diagram of a solar irradiance prediction system 100 according to embodiment 1. The solar irradiance prediction system 100 shown in Fig. 1 includes a solar irradiance prediction device 60 connected to a communication network NW such as the Internet, a performance value providing system 30, a long-term weather forecast value providing system 40, and an energy operation management system 50, all of which are also connected to the communication network NW.

[0016] The actual value providing system 30 provides actual values ​​of meteorological data such as temperature, humidity, surface pressure, sea level pressure, wind direction, wind speed, weather, and precipitation, as well as actual values ​​related to solar radiation intensity. Examples of the actual value providing system 30 include information providing systems of the Japan Meteorological Agency or Japan Meteorological Corporation.

[0017] The long-term weather forecast value providing system 40 provides forecast values ​​(weather forecast values) of weather data over a long period of several months. Examples of the long-term weather forecast value providing system 40 include the information providing systems of the Japan Meteorological Agency or Japan Weather Co., Ltd.

[0018] The energy operation management system 50 uses the solar radiation intensity predicted by the solar radiation intensity prediction system 100 to create an energy operation plan for a long period of time, up to several months in the future, that includes when to generate and store hydrogen or methane, etc., and when to purchase electricity or gas. Examples of the energy operation management system 50 include systems of electric power companies, companies in charge of energy management, and local governments.

[0019] In addition, the energy operation management system 50 can also be a control system that uses solar power generation output calculated using the solar radiation intensity predicted by the solar radiation intensity prediction system 100 and controls current and voltage using power system control equipment.

[0020] Fig. 2 is a block diagram showing the configuration of the solar radiation intensity prediction device 60. As shown in Fig. 2, the solar radiation intensity prediction device 60 includes a data acquisition unit 1, a feature generation unit 2, a learning model generation unit 3, a prediction unit 4, a solar radiation intensity statistics calculation unit 5, a transmission unit 6, and a storage unit 10.

[0021] The data acquisition unit 1 acquires weather data actual values ​​and solar radiation intensity actual values ​​from the actual value providing system 30 (FIG. 1), and acquires long-term weather data forecast values ​​from the long-term weather forecast value providing system 40.

[0022] The feature generator 2 generates feature values ​​by calculation from among the feature values ​​used in machine learning, other than the data acquired by the data acquirer 1, such as past values ​​of actual weather data and predicted weather data.

[0023] The learning model generation unit 3 uses the actual solar radiation intensity values, which are correct data, and the feature quantities to generate a learned model through machine learning for inferring the predicted value of solar radiation intensity over the long term from the feature quantities.

[0024] The prediction unit 4 uses the data stored in the storage unit 10 to predict the long-term solar radiation intensity.

[0025] The solar radiation intensity statistics calculation unit 5 calculates the statistics of solar radiation intensity for several months from the solar radiation intensity predicted by the prediction unit 4 to compensate for prediction errors that occur in long-term predictions.

[0026] The transmitter 6 transmits the solar radiation intensity statistics calculated by the solar radiation intensity statistics calculator 5 to the energy operation management system 50 .

[0027] The memory unit 10 has a database 11 that stores actual solar radiation intensity values ​​acquired by the data acquisition unit 1, a database 12 that stores features including actual weather data values ​​acquired by the data acquisition unit 1, a database 13 that stores trained models generated by the learning model generation unit 3, and a database 14 that stores predicted solar radiation intensity values.

[0028] <Operation> Below, we will explain the process in which the solar radiation intensity prediction device 60 acquires data from the actual value providing system 30 and the long-term weather forecast value providing system 40, predicts long-term solar radiation intensity, and transmits the data to the energy operation management system 50.

[0029] The processing in the solar radiation intensity prediction device 60 can be broadly divided into the following processes (1) to (3).

[0030] Process (1): Processing during learning, including generation of training data and trained models. Process (2): Processing during prediction. Process (3): Calculation of statistics. Below, processes (1) to (3) are each explained using a flowchart.

[0031] 3 is a flowchart illustrating the process (1). First, the solar radiation intensity prediction device 60 acquires, by the data acquisition unit 1, actual meteorological data values ​​and actual solar radiation intensity values ​​at the prediction target point (step S1).

[0032] That is, to use them as feature quantities, the data acquisition unit 1 acquires the weather data result values ​​and the solar radiation intensity result values ​​from the result value providing system 30 (FIG. 1) and stores them in the database 11 of the storage unit 10.

[0033] Note that instead of actual weather data values, the data acquisition unit 1 can also acquire forecast values ​​of past weather data from the long-term weather forecast value providing system 40 and store them as features in the database 12 of the storage unit 10. As a result, if there is a gap in the actual weather data values, the forecast value of the weather data can be used as a feature value for that time period. Furthermore, since the forecast value of the weather data is used when inferring the forecast value of solar radiation intensity, which will be described later, using the forecast value of past weather data instead of the actual weather data values ​​when generating a trained model enables learning that takes into account the accuracy of the forecast value of the weather data. Specifically, if the forecast value of the weather data for a certain time period tends to be underestimated compared to the actual value, learning can take that tendency into account.

[0034] Here, the data acquired by the data acquisition unit 1 may be for the last few years, for example, 2 to 5 years, of the prediction target month, but data from even further back in time may also be acquired.

[0035] Next, the data acquisition unit 1 calculates the maximum value of the solar radiation intensity for each predetermined time (t1) from the solar radiation intensity record values ​​acquired in step S1 (step S2). Here, the predetermined time (t1) is, for example, 30 minutes. In addition to the maximum value of the solar radiation intensity, the average value or minimum value of the solar radiation intensity for each predetermined time can also be calculated. The average value and minimum value of the solar radiation intensity are calculated from the solar radiation intensity record values ​​in the same way as the maximum value.

[0036] Next, the data acquisition unit 1 generates feature quantities from the actual weather data values ​​acquired in step S1 and stores them in the database 12 of the storage unit 10 (step S3). More specifically, the feature quantities (past values) are calculated by calculating average values ​​from the data of each weather item. However, the feature quantities are not limited to average values, and maximum and minimum values ​​over a 30-minute period can also be calculated and used as feature quantities.

[0037] Furthermore, it is also possible to use feature quantities generated by calculations performed by the feature quantity generation unit 2 in addition to the data acquired by the data acquisition unit 1. Examples of feature quantities generated by the feature quantity generation unit 2 include theoretical solar radiation intensity calculated using the date, time, latitude, and the like.

[0038] The theoretical solar radiation intensity is calculated using the following formula (1): Formula (1) is disclosed in Patent Document 3.

[0039]

[0040] In addition, in the formula (1), S ｔ is the theoretical solar radiation intensity, I ００ is the solar constant (= 1365 W / m ２ ), θ indicates the zenith angle, φ indicates latitude, δ indicates the solar declination, h indicates the hour angle from the solar noon, M indicates the month, DAY indicates the day, HOUR indicates the hour, and H ｎ indicates the time of noon. Also, η is a parameter for the i-th day of the year, where i=1 indicates January 1st and i=32 indicates February 1st.

[0041] FIG. 4 shows the theoretical solar radiation intensity [kW / m] at the time of noon calculated using Equation (1). ２ ] is shown in the figure, which shows the theoretical solar radiation intensity at noon from 2019 to 2021.

[0042] As shown in Figure 4, the theoretical solar radiation intensity is cyclical each year, and it can be seen that the theoretical solar radiation intensity has similar values, i.e., the same characteristics, even in January and December, which are different months. For this reason, by adding the theoretical solar radiation intensity as a feature, it becomes possible to associate "periods with similar solar radiation intensity (spring and autumn)" and "January and December are similar periods in terms of solar radiation intensity."

[0043] FIG. 5 shows learning data including feature amounts created in the processes of steps S1 to S3 and stored in the storage unit 10, and actual solar radiation intensity values ​​that are correct data corresponding to the feature amounts.

[0044] As shown in FIG. 5 , the learning data includes time, a response variable (correct answer), and explanatory variables (features). The response variable includes the “maximum solar radiation intensity,” which is the actual solar radiation intensity value, and the explanatory variables include the “average value over 30 minutes regarding the maximum value of the weather data” and the “features generated by the feature generator.”

[0045] The "30-minute average value of the maximum weather data" includes "temperature (average)" and "humidity (average)." In addition, it can also include surface pressure, sea level pressure, wind direction, wind speed, weather, and precipitation. The "features generated by the feature generator" include "theoretical solar radiation intensity."

[0046] 3, the learning model generation unit 3 generates a learned model using the learning data and a machine learning algorithm (step S4), stores the learned model in the database 13 of the storage unit 10, and ends process (1). Here, possible machine learning algorithms include, but are not limited to, random forests, regression trees, multiple regression, support vector regression, neural networks, and gradient boosting.

[0047] While the trained model described above is a model for predicting the maximum value of solar radiation intensity, if the average or minimum value of solar radiation intensity is to be predicted, a trained model will be generated separately for each. In this way, by predicting the maximum, average, and minimum values ​​of solar radiation intensity, it is possible to respond to various weather conditions.

[0048] A trained model is created for each prediction target location at a predetermined time (t1), but it is sufficient to create a trained model for the time period when solar radiation intensity is observed, for example, between 9:00 and 16:00. For example, if a trained model is created every 30 minutes between 9:00 and 16:00, a total of 14 trained models will be created.

[0049] 6 is a flowchart illustrating process (2). The solar radiation intensity prediction device 60 first obtains forecast values ​​of weather data for the target prediction time (step S11). That is, the data acquisition unit 1 obtains forecast values ​​of weather data for the target prediction time from the long-term weather forecast value providing system 40 (FIG. 1) and stores them as feature quantities in the database 12 of the storage unit 10.

[0050] When multiple forecast values ​​of weather data for the target time are provided from the long-term weather forecast value providing system 40, for example, a total of n, the data to be used as the feature may be all of them, or the average value of n of them, or the median value of n of them, but is not limited to these.

[0051] As described above, the trained model is created every predetermined time (t1). However, if the time resolution (t2) of the forecast value of the obtainable weather data is finer than the predetermined time (t1), for example, if t1 is 30 minutes and t2 is 1 minute, it is possible to take, for example, the average value over a 30-minute period. By creating a trained model every predetermined time in this way, statistical processing becomes possible. On the other hand, if the time resolution (t2) of the forecast value of the obtainable weather data is coarser than the predetermined time (t1), for example, if t1 is 30 minutes and t2 is 1 day, the forecast value of the weather data at the forecast target time, in this case, the forecast target date, S month T day, is set to a single value.

[0052] Next, feature quantities for the target prediction time are generated (step S12). As feature quantities, the predicted values ​​of the weather data acquired in step S11 can be used, or the theoretical solar radiation intensity for the target prediction time calculated by the feature quantity generator 2 can also be used. The feature quantities are generated at predetermined intervals (t1) on the target prediction day, S month, T day.

[0053] Next, the prediction unit 4 inputs the features at the prediction target time generated in step S12 into the learned model stored in the memory unit 10 and corresponding to the prediction target time, thereby performing inference using the learned model to predict the solar radiation intensity (step S13), and stores this as a solar radiation intensity predicted value in the database 14 of the memory unit 10, thereby completing process (2).

[0054] This series of processes is performed for each day of the target prediction month, month S. For example, if t1 = 0.5 hours, S = 6 months, and the time period is from 9:00 to 16:00 (total of 14 hours) each day, the predicted solar radiation intensity is obtained as follows: 30 days / month x 14 units / day = 420 units / month.

[0055] 7 is a flowchart illustrating process (3). First, the solar radiation intensity statistics calculation unit 5 acquires predicted solar radiation intensity values ​​from the database 14 in the storage unit 10 (step S21). That is, predicted solar radiation intensity values ​​are acquired for each day and hour of month S, which is the target prediction month.

[0056] Next, the solar radiation intensity statistics calculation unit 5 calculates the average value from the predicted solar radiation intensity values ​​at any time on each day of month S, and calculates the solar radiation intensity statistics (step S22). That is, in processes (1) and (2), for example, if the prediction target month is S = June and t1 = 0.5 hours, the predicted solar radiation intensity value from 12:00 to 12:30 is calculated as the average value for 30 days from the 1st to the 30th, and this average value is used as the solar radiation intensity statistics from 12:00 to 12:30. Similarly, this process is performed at each predetermined time (t1).

[0057] Next, the transmitter 6 transmits the solar radiation intensity statistics to the energy operation management system 70 (step S23), and ends the process (3).

[0058] <Effects> According to the solar radiation intensity prediction system 100 and solar radiation intensity prediction device 60 of the first embodiment described above, by transmitting solar radiation intensity statistics to the energy operation management system 70, the energy operation management system 70 can predict the amount of solar power generation and solar radiation intensity output at a certain location several months into the future, for example, three months into the future, and can predict whether surplus power or power shortage will occur several months into the future, which can be used for long-term energy operation planning.

[0059] Furthermore, by providing the solar radiation intensity statistics calculation unit 5, it is possible to compensate for errors in the long-term prediction. For example, if, in the current month of January 2024, a prediction is made for each day of April with a time resolution of 30 minutes, the prediction accuracy of the solar radiation intensity is thought to be low. On the other hand, by calculating the solar radiation intensity statistics, even if the prediction accuracy of the solar radiation intensity for each day is low, the prediction accuracy can be improved as long as it is accurate to the extent that the solar radiation intensity in April 2024 is higher than the statistics of the solar radiation intensity for April 2023 and 2022. In other words, in long-term prediction, if it is possible to grasp the statistics of a certain month several months in the future rather than a predicted value for each day with a fine time resolution several months in the future, this can be used for energy management such as peak shifting.

[0060] In addition, users of the solar radiation intensity prediction device 60, such as electric power companies, companies in charge of energy management, and local governments, can generate trained models themselves and can generate trained models using data that they do not want to release to the outside.

[0061] <Embodiment 2> <Device configuration> In the solar radiation intensity prediction device 60 of embodiment 1 shown in Figure 2, both the generation of a trained model and the prediction using the trained model are performed by a single device that cannot be separated, but the generation of a trained model and the prediction using the trained model can also be performed by separate devices that can be separated.

[0062] 8 is a block diagram showing the configuration of a solar radiation intensity prediction device 60A according to a second embodiment, in which generation of a trained model and prediction using the trained model are performed by separate devices. Note that the solar radiation intensity prediction system 200 according to the second embodiment is the same as the solar radiation intensity prediction system 100 according to the first embodiment shown in FIG. 1, except that the solar radiation intensity prediction device 60 is replaced with the solar radiation intensity prediction device 60A, and therefore a duplicated description will be omitted.

[0063] As shown in Figure 8, the solar radiation intensity prediction device 60A is composed of a learning device 61 (first device) that generates a trained model and a prediction device 62 (second device) that makes predictions using the trained model, and the prediction device 62 predicts solar radiation intensity using the trained model generated by the learning device 61 and data from the long-term weather forecast value providing system 40.

[0064] The learning device 61 includes a data acquisition unit 1, a feature generation unit 2, a learning model generation unit 3, and a storage unit 10.

[0065] The data acquisition unit 1 acquires weather data actual values ​​and solar radiation intensity actual values ​​from the actual value providing system 30 , and acquires long-term weather data forecast values ​​from the long-term weather forecast value providing system 40 .

[0066] The feature generation unit 2 generates feature values ​​by calculation from among the feature values ​​used in machine learning, other than the data acquired by the data acquisition unit 1, for example, past values ​​of the actual weather data and the predicted weather data.

[0067] The learning model generation unit 3 uses the actual solar radiation intensity values, which are correct data, and the feature quantities to generate a learned model through machine learning for inferring the predicted value of solar radiation intensity over the long term from the feature quantities.

[0068] The memory unit 10 has a database 11 that stores the actual weather data values ​​and actual solar radiation intensity values ​​acquired by the data acquisition unit 1, a database 12 that stores feature values, and a database 13 that stores the trained model generated by the learning model generation unit 3.

[0069] The prediction device 62 includes a data acquisition unit 21 , a feature generation unit 22 , a prediction unit 24 , a solar radiation intensity statistics calculation unit 25 , a transmission unit 26 , and a storage unit 20 .

[0070] The data acquisition unit 21 acquires forecast values ​​of weather data for the target forecast time from the long-term weather forecast value providing system 40 and stores them as feature quantities in the database 32 of the storage unit 20 .

[0071] The feature generator 22 generates feature values ​​by calculation from among the feature values ​​used in machine learning, other than the data acquired by the data acquirer 1, such as past values ​​of actual weather data and forecasted weather data.

[0072] The prediction unit 24 uses the data stored in the storage unit 20 to predict the long-term solar radiation intensity.

[0073] The solar radiation intensity statistics calculation unit 25 calculates the statistics of solar radiation intensity for several months from the solar radiation intensity predicted by the prediction unit 24 to compensate for prediction errors that occur in long-term predictions.

[0074] The transmitting unit 26 transmits the solar radiation intensity statistics calculated by the solar radiation intensity statistics calculating unit 25 to the energy operation management system 50 .

[0075] The memory unit 20 has a database 32 that stores the predicted values ​​of weather data for the prediction time acquired by the data acquisition unit 21 from the weather forecast value providing system 40, a database 33 that stores the learned model generated by the learning model generation unit 3, and a database 34 that stores the predicted solar radiation intensity values.

[0076] <Operation> Below, we will explain the process in which the solar radiation intensity prediction device 60A acquires data from the actual value providing system 30 and the long-term weather forecast value providing system 40, predicts long-term solar radiation intensity, and transmits the data to the energy operation management system 50.

[0077] The processing in the solar radiation intensity prediction device 60A can be broadly divided into the following processes (11) to (13).

[0078] Process (11): Processing in the learning device 61, including generation of learning data and a learned model. Process (12): Prediction processing in the prediction device 62. Process (13): Calculation processing of statistics in the prediction device 62. Processes (11) to (13) will be explained using the flowcharts shown in Figures 3, 6, and 7, respectively.

[0079] 3 is a flowchart illustrating the process (11). First, the learning device 61 acquires the actual weather data values ​​and the actual solar radiation intensity values ​​at the prediction target point using the data acquisition unit 1 (step S1).

[0080] That is, to use them as feature quantities, the data acquisition unit 1 acquires the weather data result values ​​and the solar radiation intensity result values ​​from the result value providing system 30 (FIG. 1) and stores them in the database 11 of the storage unit 10.

[0081] In addition, instead of actual weather data values, the data acquisition unit 1 can also acquire forecast values ​​of past weather data from the long-term weather forecast value providing system 40 and store them as features in the database 12 of the storage unit 10.

[0082] Next, the data acquisition unit 1 calculates the maximum value of the solar radiation intensity for each predetermined time (t1) from the solar radiation intensity record values ​​acquired in step S1 (step S2), where the predetermined time (t1) is, for example, 30 minutes.

[0083] Next, the data acquisition unit 1 generates feature quantities from the weather data performance values ​​acquired in step S1, and stores the feature quantities in the database 12 of the storage unit 10 (step S3).

[0084] Furthermore, in addition to the data acquired by the data acquisition unit 1, it is also possible to use feature quantities generated by calculations performed by the feature quantity generation unit 2. Examples of feature quantities generated by the feature quantity generation unit 2 include theoretical solar radiation intensity calculated using Equation (1).

[0085] Next, the learning model generation unit 3 generates a learned model using the learning data and the machine learning algorithm (step S4), stores the learned model in the database 13 of the storage unit 10, and ends process (1).

[0086] 6 is a flowchart illustrating process (12). The prediction device 62 first obtains forecast values ​​of weather data for the target prediction time (step S11). That is, the data acquisition unit 21 obtains forecast values ​​of weather data for the target prediction time from the long-term weather forecast value providing system 40, and stores them as feature values ​​in the database 32 of the storage unit 20.

[0087] Next, a feature value for the target prediction time is generated (step S12). As the feature value, the predicted value of the weather data acquired in step S11 can be used, or the theoretical solar radiation intensity for the target prediction time calculated by the feature value generator 22 can also be used.

[0088] Next, the prediction unit 24 predicts the solar radiation intensity by inputting the features at the prediction target time generated in step S12 into the learned model stored in the memory unit 20 and corresponding to the prediction target time (step S13), and stores the predicted solar radiation intensity value in the database 34 of the memory unit 20, thereby completing process (2).

[0089] 7 is a flowchart illustrating process (13). First, the solar radiation intensity statistics calculation unit 5 acquires predicted solar radiation intensity values ​​from the database 34 of the storage unit 20 (step S21). That is, predicted solar radiation intensity values ​​are acquired for each day and hour of month S, which is the prediction target month.

[0090] Next, the solar radiation intensity statistics calculation unit 5 calculates the average value from the predicted solar radiation intensity values ​​at any time on each day of the S month, and calculates the solar radiation intensity statistics (step S22).

[0091] Next, the transmitter 26 transmits the solar radiation intensity statistics to the energy operation management system 70 (step S23), and ends the process (13).

[0092] <Effects> In the solar radiation intensity prediction device 60A of the second embodiment described above, the learning device 61 and the prediction device 62 can be physically separated, so that each can be owned by a separate entity, for example, the manufacturer of the solar radiation intensity prediction device 60A can own the learning device 61, which has a low update frequency and a high processing load, and the user who operates the energy operation management system 50 can own the prediction device 62.

[0093] For this reason, a trained model is generated by a learning device 61 owned by a manufacturer, and the generated trained model is incorporated into a prediction device 62 owned by a user. In this case, the prediction device 62 does not have a learning model generation unit 3 that requires a large amount of processing, and therefore can be a cheaper device than the solar radiation intensity prediction device 60 of the first embodiment.

[0094] Furthermore, the solar radiation intensity prediction device 60 of the first embodiment is owned by the user, and a trained model is generated for each user. On the other hand, in the second embodiment, if the manufacturer owns the learning device, the manufacturer can perform the learning collectively, which previously was performed for each user, thereby reducing the burden of learning on the user.

[0095] The trained model can be provided to the user, for example, by using the learning device 61 as a server computer, which generates and returns a trained model when the user inputs performance value information via a communication network. In this case, more training data is available than when a trained model is generated for each user, making it possible to generate a trained model with high accuracy.

[0096] The learning device 61 is not limited to being owned by the manufacturer, but may be owned by a user other than the user who owns the prediction device 62.

[0097] The solar radiation intensity prediction device 60A of the second embodiment described above is equipped with a learning device 61 and a prediction device 62, but if a trained model such as that generated by the learning device 61 can be obtained from the outside, the solar radiation intensity prediction device can be configured with only the prediction device 62, making it possible to make a cheaper device.

[0098] <Hardware Configuration> Note that each of the components of the solar radiation intensity prediction devices 60 and 60A according to the first and second embodiments described above can be configured using a computer and implemented by the computer executing a program. That is, the solar radiation intensity prediction devices 60 and 60A are implemented by, for example, a processing circuit 500 shown in Fig. 9. A processor such as a CPU (Central Processing Unit) or a DSP (Digital Signal Processor) is applied to the processing circuit 500, and the functions of each part are implemented by executing a program stored in a storage device.

[0099] Dedicated hardware may be applied to the processing circuit 500. When the processing circuit 500 is dedicated hardware, the processing circuit 500 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination thereof.

[0100] In the solar radiation intensity prediction devices 60 and 60A, the functions of the components may be realized by individual processing circuits, or the functions may be realized together by a single processing circuit.

[0101] 10 also shows a hardware configuration in the case where the processing circuit 500 is configured using a processor. In this case, the functions of each part of the solar radiation intensity prediction devices 60 and 60A are realized by a combination of software, etc. (software, firmware, or software and firmware). The software, etc. is written as a program and stored in memory 520. The processor 510 functioning as the processing circuit 500 realizes the functions of each part by reading and executing the program stored in memory 520 (storage device). In other words, it can be said that this program causes a computer to execute the procedures and methods of operation of the components of the solar radiation intensity prediction devices 60 and 60A.

[0102] Here, the memory 520 may be, for example, a non-volatile or volatile semiconductor memory such as RAM, ROM, flash memory, EPROM (Erasable Programmable Read Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), HDD (Hard Disk Drive), magnetic disk, flexible disk, optical disk, compact disk, mini disk, DVD (Digital Versatile Disc) and its drive device, or any storage medium that will be used in the future.

[0103] The above describes a configuration in which the functions of each component of the solar irradiance prediction devices 60 and 60A are realized by either hardware or software, etc. However, this is not limited to this, and the solar irradiance prediction devices 60 and 60A may be configured such that some of the components are realized by dedicated hardware and other components are realized by software, etc. For example, some of the components may be realized by the processing circuit 500 as dedicated hardware, and other components may be realized by the processing circuit 500 as the processor 510 reading and executing a program stored in the memory 520.

[0104] As described above, the solar radiation intensity prediction devices 60 and 60A can realize the above-described functions by hardware, software, or a combination of these.

[0105] Although the present disclosure has been described in detail, the above description is illustrative in all respects and does not limit the disclosure thereto. It is understood that countless variations not illustrated can be envisioned without departing from the scope of the present disclosure.

[0106] It should be noted that, within the scope of the present disclosure, the embodiments can be freely combined, modified, or omitted as appropriate.

Claims

1. A solar radiation intensity prediction device that predicts solar radiation intensity over a long period of one month or more using predicted values ​​of meteorological data, actual solar radiation intensity values, and actual weather data values, comprising: a learning model generation unit that generates a trained model by machine learning using the actual solar radiation intensity values, which are correct data, and feature values, to infer the predicted value of the solar radiation intensity over the long period; a feature generation unit that generates calculated feature values ​​among the feature values ​​other than past values ​​of the actual weather data values ​​and the predicted value of the meteorological data; a prediction unit that predicts the solar radiation intensity over the long period to generate a solar radiation intensity predicted value; a memory unit that stores the actual weather data values, the actual solar radiation intensity values, the predicted solar radiation intensity values, the feature values, and the trained model; a solar radiation intensity statistics calculation unit that calculates statistics of the solar radiation intensity from the predicted solar radiation intensity values ​​to compensate for prediction errors that occur in long-term predictions; and a transmission unit that transmits the statistics, wherein the prediction unit: A solar radiation intensity prediction device that inputs the feature amount into the trained model, performs inference using the trained model, and obtains the solar radiation intensity prediction value.

2. A solar radiation intensity prediction device that predicts solar radiation intensity over a long period of one month or more using predicted values ​​of meteorological data, actual solar radiation intensity values, and actual weather data values, the device comprising a first device and a second device that are separable from each other, wherein the first device comprises: a learning model generation unit that generates a trained model by machine learning using the actual solar radiation intensity values, which are correct data, and features to infer predicted values ​​of solar radiation intensity over the long period; a feature generation unit that generates calculated features from the feature values ​​other than past values ​​of the actual weather data values ​​and the predicted weather data values; and a storage unit that stores the actual weather data values, the actual solar radiation intensity values, the features, and the trained model, and the second device comprises: a prediction unit that predicts the solar radiation intensity over the long period and sets it as a solar radiation intensity predicted value; another feature generation unit that generates the calculated features from the feature values; and another storage unit that acquires and stores the features, the predicted solar radiation intensity values, and the trained model from the storage unit. A solar radiation intensity prediction device comprising: a solar radiation intensity statistics calculation unit that calculates statistics of the solar radiation intensity from the solar radiation intensity predicted value to compensate for prediction errors that occur during long-term prediction; and a transmission unit that transmits the statistics, wherein the prediction unit inputs the features into the trained model, thereby performing inference using the trained model to obtain the solar radiation intensity predicted value.

3. A solar radiation intensity prediction device that predicts solar radiation intensity over a long period of one month or more using predicted values ​​of meteorological data, actual solar radiation intensity values, and actual weather data values, the solar radiation intensity prediction device comprising: a memory unit that acquires and stores a trained model for inferring the predicted value of solar radiation intensity over the long period, the trained model being generated by machine learning using the actual solar radiation intensity values, which are correct data, and features including past values ​​of the actual weather data and the predicted values ​​of the weather data; a prediction unit that predicts the solar radiation intensity over the long period to generate a solar radiation intensity predicted value; another feature generation unit that generates features by calculation; a solar radiation intensity statistics calculation unit that calculates statistics of the solar radiation intensity from the predicted solar radiation intensity values ​​to compensate for prediction errors that occur in long-term predictions; and a transmission unit that transmits the statistics, wherein the memory unit also stores the features and the predicted solar radiation intensity values, and the prediction unit inputs the features into the trained model to perform inference using the trained model to obtain the predicted solar radiation intensity value.

4. A solar radiation intensity prediction device according to claim 1 or claim 3, wherein the feature generation unit generates theoretical solar radiation intensity for each month, day and hour as the calculated feature.

5. The solar radiation intensity prediction device according to claim 2, wherein the feature generation unit and the other feature generation unit generate theoretical solar radiation intensity for each month, day, and hour as the calculated feature.

6. A solar radiation intensity prediction device according to claim 1 or claim 2, wherein the learning model generation unit generates the learned model at predetermined time intervals.

7. A solar radiation intensity prediction device as described in claim 1 or claim 2, wherein the learning model generation unit generates the learned models that predict the maximum, average, and minimum values ​​of the solar radiation intensity for each specified time period.

8. A solar radiation intensity prediction system comprising: a solar radiation intensity prediction device according to any one of claims 1 to 3; an actual value providing system that provides the actual solar radiation intensity values ​​and the actual weather data values; a weather forecast value providing system that provides forecast values ​​of the weather data; and an energy operation management system that receives the statistics transmitted from the transmission unit.

Images