JP7128106B2 - PV output prediction support device, PV output prediction device, PV output prediction support method, and PV output prediction support program - Google Patents

Description

Application of Article 30, Paragraph 2 of the Patent Act 2018 The Institute of Electrical Engineers of Japan Power and Energy Division Conference Proceedings Published August 31, 2018 The Institute of Electrical Engineers of Japan Power and Energy Division Conference Date September 12, 2018

The present invention relates to a PV output prediction support device, a PV output prediction device, a PV output prediction support method, and a PV output prediction support program.

It is expected that the number of consumers who use energy equipment such as photovoltaics (PV) and storage batteries will increase in the future toward the realization of a low-carbon society and the efficient use of energy. In particular, after the enforcement of FIT (Feed In Tariff), the introduction of photovoltaic systems (photovoltaics) to power distribution systems is rapidly progressing.

If a large amount of photovoltaic power generation is introduced into the power system of an electric power company, it is feared that the power supply and demand operation of the electric power company will be greatly affected by the output fluctuation. Under these circumstances, in order to maintain stable supply and demand, it is necessary to accurately predict the output of photovoltaic power generation in areas of various sizes, from the wide area level of the entire power system to the district level of each distribution substation. are required to do so. Below, the output of photovoltaic power generation is called "PV output."

The amount of solar radiation used to predict the PV output fluctuates due to daily weather changes. Therefore, the amount of solar radiation can be a large error factor in the prediction of PV output. For example, temporal and spatial prediction errors for the front may cause large deviations in the amount of solar radiation and the timing of its changes, and may have a large impact on supply and demand management.

In order to reduce such prediction errors, techniques have been proposed that attempt to improve prediction accuracy by combining a plurality of PV output prediction models. Since the PV output can be obtained if the actual amount of solar radiation is known, the prediction model for the PV output can also be considered as a prediction model for the actual amount of solar radiation. With this technology, it is possible to make predictions in consideration of output fluctuations such as upward and downward fluctuations in PV output. In addition, it is possible to evaluate the reliability of the prediction results from the variation of each prediction model and the prediction results from the past to the present.

Here, as weather information prediction technology, satellite images are grouped based on associated weather information and feature information, similar groups are identified from the feature information of satellite images to be predicted, and weather information associated with the group is identified. There is a conventional technology for predicting future weather by using (Patent Document 1). Further, as a PV output prediction technology, there is a conventional technology for estimating the current PV output using a predetermined conversion formula for data on days similar to the prediction target date (Patent Document 2). In addition, the weather record and weather forecast for each time period in the past are acquired, classified into similar time periods, and based on the actual power generation amount and the amount of extra-atmospheric solar radiation in the similar time period, the prediction target in the type of actual weather record There is a conventional technique for obtaining a prediction formula for calculating a predicted amount of power generation from the amount of extra-atmospheric solar radiation in a time period (Patent Document 3).

JP 2016-205930 A JP 2014-200167 A WO2017/026010

However, with conventional technology that handles multiple forecast models, when there is a large variation in forecast results among PV output forecast models, it is important to select which model for supply and demand operation. The selection is made based on the experience of Selection based on the operator's experience may result in a decrease in the accuracy of the PV output prediction result due to a selection error or the like.

In addition, none of the above-described conventional techniques considers handling of a plurality of PV output prediction models, and it is difficult to improve the accuracy of PV output prediction by appropriate model selection.

The disclosed technology has been made in view of the above, and provides a PV output prediction support device, a PV output prediction device, a PV output prediction support method, and a PV output prediction support program that improve the accuracy of PV output prediction. for the purpose.

In one aspect of the PV output prediction support device, the PV output prediction device, the PV output prediction support method, and the PV output prediction support program disclosed in the present application, the storage unit stores information on the distribution of solar radiation for past days in the target area, the past It stores the measurement result of the photovoltaic power generation output of the day and each prediction result in the past days when using a plurality of prediction models used for the prediction of the photovoltaic power generation output. The extracting unit acquires information on the solar radiation distribution of the prediction target day, and extracts similar days having the solar radiation distribution similar to the prediction target day from among the selection target days included in a predetermined period of the past days. do. A prediction error acquisition unit acquires a prediction error for each prediction model on the similar date extracted by the extraction unit based on the measurement result and each prediction result. The notification unit notifies each of the prediction errors acquired by the prediction error acquisition unit as a prediction error of each of the prediction models for predicting the photovoltaic power generation output on the prediction target day.

In one aspect, the present invention can improve the accuracy of PV output prediction.

FIG. 1 is an overall configuration diagram of a PV output prediction support system. FIG. 2 is a block diagram of the PV output prediction device according to the first embodiment. FIG. 3 is a diagram showing classification results based on clustering. FIG. 4 is a diagram showing an example of error evaluation results of each prediction model. FIG. 5 is a diagram summarizing the evaluation results of each group. FIG. 6 is a diagram for explaining selection target dates. FIG. 7 is a diagram for explaining group determination of selection target dates. FIG. 8 is a diagram for explaining calculation of the RMSE between the distribution of solar radiation on the prediction target day and the distribution of solar radiation for each group. FIG. 9 is a diagram for explaining the calculation of the RMSE between the solar radiation distribution on the prediction target day and the solar radiation distribution on the comparison target day. FIG. 10 is a diagram for explaining the calculation of the RMSE between the solar radiation amount difference distribution of the prediction target day and the solar radiation amount difference distribution of the comparison target day. FIG. 11 is a diagram for explaining the calculation of the standard deviation of the solar radiation amount on the prediction target day using the solar radiation distribution at the prediction start time of the prediction target day and the average value of the solar radiation amount on the prediction target day. FIG. 12 is a diagram for explaining a method of determining cumulus clouds during analysis using standard deviations. FIG. 13 is a diagram showing observation points of the meteorological office in the target area. FIG. 14A is a diagram showing the first page of a web page that provides information on a solar radiation amount prediction result. FIG. 14B is a diagram showing the solar radiation amount prediction result for the first page of the web page. FIG. 15 is a diagram showing the second page of the web page that provides information on PV output prediction results. FIG. 16 is a flowchart of processing for notifying the prediction error of each prediction model by the PV output prediction device according to the first embodiment. FIG. 17 is a flowchart of a similar date extraction process. FIG. 18 is a block diagram of a PV output prediction device according to the second embodiment. FIG. 19 is a diagram illustrating an example of a web page provided by the PV output prediction device according to the third embodiment; FIG. 20 is a diagram showing an example of another web page provided by the PV output prediction device according to the third embodiment. FIG. 21 is a hardware configuration diagram of a PV output prediction device.

Exemplary embodiments of a PV output prediction support device, a PV output prediction device, a PV output prediction support method, and a PV output prediction support program disclosed in the present application will be described below in detail with reference to the drawings. Note that the PV output prediction support device, the PV output prediction device, the PV output prediction support method, and the PV output prediction support program disclosed in the present application are not limited to the following embodiments.

FIG. 1 is an overall configuration diagram of a PV output prediction support system. As shown in FIG. 1, the PV output prediction support system according to this embodiment includes a PV output prediction device 1, a weather data delivery system 2, and a terminal device 3. FIG.

The meteorological data distribution system 2 acquires meteorological satellite images captured by Himawari-8, for example. Then, the weather data distribution system 2 accumulates the acquired weather satellite images.

The PV output prediction device 1 acquires weather satellite images from the weather data distribution system 2 . Then, the PV output prediction device 1 calculates a solar radiation distribution from the obtained weather satellite image, and uses the calculated solar radiation distribution to generate information for supporting selection of a prediction model used for PV output prediction. After that, the PV output prediction device 1 provides the terminal device 3 with information that supports the selection of the generated prediction model. The PV output prediction device 1 provides, for example, information that assists the selection of a prediction model at intervals of 30 minutes.

Here, the PV output prediction device 1 does not have to directly send the information supporting the selection of the prediction model to the terminal device 3 . As a method of providing information to support the selection of a prediction model, for example, the PV output prediction device 1 browses a page containing prediction support information for supporting the selection of a prediction model on a web server accessible from the terminal device 3. can be made possible. In this case, the PV output prediction device 1 updates the information on the web page every 30 minutes, for example.

The terminal device 3 is installed at an electric power company or the like. The terminal device 3 acquires prediction support information from the PV output prediction device 1 . Then, the terminal device 3 provides the user with the prediction support information by outputting it to an output device such as a monitor. The user confirms the prediction support information on a monitor or the like and determines which of the plurality of prediction models is appropriate when performing prediction at the present time. The user then selects the optimum prediction model to predict the current PV output.

Creation of selection support information by the PV output prediction device 1 will be described in detail below. In this embodiment, the PV output prediction device 1 predicts the PV output in 30-minute intervals from the prediction start time until 3 hours later, that is, in 6 time zones. Hereinafter, the day on which the PV output prediction device 1 makes a prediction will be referred to as a "prediction target day". That is, the prediction target day indicates the day on which the PV output prediction device 1 makes the prediction. Also, the time at which prediction is started is referred to as "prediction start time", and the time period to be subjected to prediction is referred to as "prediction target time period". In the present embodiment, the prediction target time zone includes a time zone every 30 minutes with a prediction start time of 9:30 and a prediction end time of 12:30. Also, in the following description, a case where three different prediction models are used as prediction models will be described. Each prediction model is called first to third prediction models. Furthermore, an area targeted for PV output prediction is referred to as a “target area”. In this embodiment, as an example, the target area is assumed to be an area including mainland Kyushu.

FIG. 2 is a block diagram of the PV output prediction device according to the first embodiment. As shown in FIG. 2, the PV output prediction device 1 includes a solar radiation distribution estimation unit 11, a storage unit 12, a classification unit 13, a prediction error calculation unit 14, an extraction unit 15, a prediction error acquisition unit 16, a notification unit 17, and a PV It has an output prediction unit 18 .

The solar radiation distribution estimation unit 11 acquires daily weather satellite images from the weather data delivery system 2 . In the present embodiment, hourly weather satellite images are acquired for the prediction target date, the days of the current year up to the prediction target date, and all the days of the previous year and the year before last. Hereinafter, the days of the current year up to the prediction target date and all the days of the previous year and the year before last are referred to as "past days".

Next, the solar radiation distribution estimating unit 11 obtains the cloud albedo for each predetermined unit area of the target area from the meteorological satellite images for each time zone of each day. Here, the predetermined unit area is, for example, 1 km ² . Furthermore, the solar radiation distribution estimator 11 calculates the ground surface albedo using the obtained cloud albedo. After that, the solar radiation distribution estimating unit 11 calculates the calculated cloud albedo and ground surface albedo, the correction coefficient for the atmosphere during fine weather, the correction coefficient for considering the type of clouds, the correction for considering the influence of air mass The global solar radiation is calculated using the coefficient and the horizontal solar radiation incident on the horizontal unit area of the top of the atmosphere. Then, the solar radiation distribution estimating unit 11 uses a summary of the global solar radiation in the target area as the solar radiation distribution to be used for PV output prediction. After that, the solar radiation distribution estimating unit 11 stores the solar radiation distribution in each time zone of each past day in the solar radiation distribution database 122 .

Moreover, the prediction is continuously performed at intervals of 10 minutes, and the solar radiation distribution estimation unit 11 also acquires the weather satellite image at the prediction start time. Therefore, the solar radiation distribution estimation unit 11 also stores the solar radiation distribution at the prediction start time of the prediction target day in the solar radiation distribution database 122 .

Here, the solar radiation distribution estimating unit 11 periodically estimates the solar radiation distribution in advance, and at the timing of performing the PV output prediction process, newly calculates the solar radiation amount for the date and time when the solar radiation distribution is not estimated. A distribution may be estimated and added to the solar radiation amount distribution database 122 .

The storage unit 12 has relative humidity data 121 , a solar radiation distribution database 122 , PV output performance data 123 and PV output prediction result data 124 . The solar radiation distribution database 122 registers the solar radiation distribution by the solar radiation distribution estimator 11 .

The relative humidity data 121 is registered with the relative humidity for each time period on the prediction target day and the past day at a predetermined point in the target area obtained from the weather station of the Japan Meteorological Agency.

As the PV output performance data 123, information on the measured value of the PV output in the target area for each time zone on the past day is registered. The PV output prediction result data 124 stores the prediction result of the PV output prediction performed by the PV output prediction unit 18, which will be described later. That is, in the PV output prediction result data 124, information of predicted values for each prediction model of PV output in the target area for each time zone of the past day is registered. Then, in the PV output prediction result data 124, three prediction results are registered for each time zone of each past day. The PV output prediction result data 124 also stores the prediction result of the prediction date. The PV output prediction result data 124 also registers the prediction result of the solar radiation distribution.

The classification unit 13 acquires the solar radiation distribution for each prediction target time zone of the past day from the solar radiation distribution database 122 . Next, the classifying unit 13 classifies the solar radiation distribution for each time zone of the prediction target day of the past day into 16 groups using hierarchical clustering without learning. A specific clustering method will be described below.

The classification unit 13 normalizes each solar radiation amount with the extra-atmospheric solar radiation amount. Next, the classification unit 13 smoothes the data in each time period. Next, the classification unit 13 round-robinly calculates the correlation coefficient σ of the spatial distribution of the amount of solar radiation in the same time zone for each past day. Next, the classification unit 13 subtracts the value obtained by squaring the correlation coefficient σ from 1, that is, the value of σ of “1−σ ² ” represented as a threshold in increments of 0.001 from 1.0 as data. Let D be the index of the closeness between the two. Next, the classification unit 13 combines pairs with high correlation coefficients. Next, the classification|category part 13 calculates an average solar radiation amount by the pair which combined. The classification unit 13 repeats pair creation until the correlation coefficient does not exceed the threshold. Here, the threshold is, for example, σ=0.3 (D=0.91).

By this clustering, the classification unit 13 can classify the past days into 16 groups as indicated by classification 131 in FIG. FIG. 3 is a diagram showing classification results based on clustering. The 16 group classifications are standardized solar radiation distributions. Here, in FIG. 3, the solar radiation amount is obtained by multiplying the normalized solar radiation amount for each unit area of each group by the extra-atmospheric solar radiation amount, and is created by color-coding according to the solar radiation amount. represents each group. In the figure, dark-colored parts are areas with many clouds, and light-colored parts are areas with few clouds. Next, as indicated by classification 132, the classification unit 13 rearranges the order of the created groups so that the number of clouds increases. As a result, the classification unit 13 determines groups #1 to #16 shown in the classification 132 .

That is, the distribution obtained by multiplying the normalized solar radiation amount distribution of each group #1 to #16 generated by the classification unit 13 by the extra-atmospheric solar radiation amount on a specific date and time is shown in each image of the classification 132 in FIG. It corresponds to the solar radiation distribution. Hereinafter, the solar radiation distribution obtained by multiplying the normalized solar radiation distribution of groups #1 to #16 by the extra-atmospheric solar radiation on a specific date and time is simply referred to as "group #1 to #16 solar radiation distribution".

After that, the classification unit 13 outputs the classification result to the prediction error calculation unit 14 and the selection target date narrowing unit 151 of the extraction unit 15 . The classification result includes information on past days belonging to each group #1 to #16. Further, the classification unit 13 outputs the respective representative solar radiation amount distributions of the groups #1 to #16 to the selection target date narrowing unit 151 .

The prediction error calculation unit 14 receives the input of the classification result from the classification unit 13 . Next, the prediction error calculation unit 14 acquires the PV output measurement result for each prediction target time period on each past day from the PV output performance data 123 . The prediction error calculator 14 also acquires the PV output prediction result for each prediction target time slot of each past day from the PV output prediction result data 124 .

Next, the prediction error calculator 14 performs error evaluation of each prediction model in each group using RMSE (Root Mean Squared Error), which is the square root of the mean squared error. That is, the prediction error calculator 14 performs error evaluation using the following formula (1). Here, the target period used for the error evaluation is three months, which is one month before and after the target month including the prediction target date. Here, the reason why the target period is narrowed down to three months, which is one month before and after the target month, is that if the day is not close to the forecast target date, there will be a large difference in the forecast results due to differences in seasons and other factors. It's for. RMSE is an example of an "error evaluation function."

Here, PV _model is a prediction result of PV output. Also, PV _obs is the measurement result of the PV output. Here, the square of the difference between the PV _model of the number of days in the target period belonging to the group for error evaluation and the PV _obs is calculated, and the sum of the calculation results is obtained. Also, N is the number of days within the target period belonging to the group for error evaluation.

That is, the prediction error calculation unit 14 divides the sum of the squared differences between the PV output prediction result and the PV output measurement result for each day in the target period by the number of days in the target period, and then takes the square root of the value. , the RMSE of the predicted PV output and the measured PV output. Then, the prediction error calculator 14 creates a prediction error graph representing the error evaluation result of each prediction model using the calculated RMSE. However, in the present embodiment, the prediction error calculation unit 14 does not perform error evaluation when the number of samples for which error evaluation is performed is three or less, since it is difficult to perform appropriate error evaluation.

The prediction error calculation unit 14 performs error evaluation of each prediction model described above for each prediction target time zone. That is, in the present embodiment, error evaluation is performed for each of the prediction models for six time periods from 9:30, which is the prediction start time, to 12:30, which is 180 minutes ahead. The prediction error calculator 14 creates a prediction error graph illustrated in FIG.

FIG. 4 is a diagram showing an example of error evaluation results of each prediction model. Prediction error graphs 41 and 42 represent error evaluation results in any two of groups #1 to #16. In the prediction error graphs 41 and 42, the vertical axis represents the REMS value and the horizontal axis represents the time period. A line segment 21 in the prediction error graphs 41 and 42 represents the error of each time zone of the first prediction model. A line segment 22 represents the error of each time period of the second prediction model. A line segment 23 represents the error of each time zone of the third prediction model. A smaller value of RMSE indicates a smaller error. Therefore, in the prediction error graph 41, the second prediction model represented by the line segment 22 has the smallest error. Also, in the prediction error graph 42, the third model represented by the line segment 23 has the smallest error.

Furthermore, the prediction error calculator 14 writes a cluster map indicating the prediction start time and the classification number in the prediction error graphs 41 and 42 . In addition, the prediction error calculator 14 describes the number of days within the target period belonging to the group represented by the prediction error graph 41, that is, the number of samples used for error evaluation. Here, in the present embodiment, the prediction error calculation unit 14 regards the case where the number of samples is less than 10 as a case where the number of samples is small and the accuracy of the evaluation result of the error evaluation is likely to be low. Highlighting 24 for notification is performed.

FIG. 5 is a diagram summarizing the evaluation results of each group. The prediction error calculator 14 performs error evaluation for each prediction model described above for each group #1 to #16, and generates prediction error graphs 201 to 216. FIG. Here, prediction error graphs 201 to 216 are evaluation results for groups #1 to #16, respectively. As shown in the prediction error graphs 203, 207, 210 and 212-216, error evaluation is not performed for groups #3, #7, #10 and #12-#16 because the number of samples is small. In this way, there are groups that do not occur or rarely occur depending on the season of the prediction target day.

The extraction unit 15 extracts similar past days with similar weather conditions such as solar radiation distribution, cloud distribution, and relative humidity to the day to be predicted. The extraction unit 15 includes a selection target date narrowing unit 151 , a first difference calculation unit 152 , a second difference calculation unit 153 , a relative humidity difference calculation unit 154 , a cloud distribution calculation unit 155 and a similar date extraction unit 156 .

As shown in FIG. 6, the selection target date narrowing unit 151 specifies 45 days before and after the same day as the prediction target date and 45 days before the prediction target date from past days as selection target dates. FIG. 6 is a diagram for explaining selection target dates. Day 31 is the prediction target day. The 45 days before and after the same day as the prediction target date and the 45 days before the prediction target date correspond to an example of the “predetermined period”. Day 32 is the same day as the prediction target day one year ago. Also, day 33 is the same day as the prediction target day two years ago. The selection date narrowing unit 151 identifies days included in periods D1, D2, and D3 as selection target dates.

Next, the selection target date narrowing unit 151 acquires the solar radiation distribution at the prediction start time of the selection target date from the solar radiation distribution database 122 . The selection target date narrowing unit 151 also receives an input of the classification result from the classification unit 13 . Then, the selection target day narrowing unit 151 determines to which group #1 to #16 the solar radiation amount distribution at the prediction start time of each selection target date belongs. Here, the group to which the solar radiation distribution at the prediction start time on a specific day belongs is referred to as a "specific day group".

For example, the selection target date group determination by the selection target date narrowing unit 151 will be described with reference to FIG. 7 as an example. FIG. 7 is a diagram for explaining group determination of selection target dates. A case will be described where eight solar radiation distributions 51 to 58 shown in FIG. 7 are included in the solar radiation distribution at the prediction start time of the selected day. The selection target date narrowing unit 151 determines that the solar radiation distribution 51 belongs to group #4 based on the classification results obtained from the classification unit 13 . Further, the selection target date narrowing unit 151 determines that the solar radiation distribution 52 belongs to group #7. Further, the selection target date narrowing unit 151 determines that the solar radiation distribution 53 belongs to group #16. Further, the selection target date narrowing unit 151 determines that the solar radiation distribution 54 belongs to group #6. Further, the selection target date narrowing unit 151 determines that the solar radiation distribution 55 belongs to group #8. Further, the selection target date narrowing unit 151 determines that the solar radiation distribution 56 belongs to group #16. Further, the selection target date narrowing unit 151 determines that the solar radiation distribution 57 belongs to group #15. Further, the selection target date narrowing unit 151 determines that the solar radiation distribution 58 belongs to group #15. In this way, the selection target day narrowing unit 151 repeats the determination for the solar radiation amount distribution of the prediction start time for all selection target days.

Next, the selection target date narrowing unit 151 acquires the solar radiation distribution at the prediction start time of the prediction target date from the solar radiation distribution database 122 . Further, the selection target date narrowing unit 151 acquires the solar radiation amount distribution of each of the groups #1 to #16 from the solar radiation amount distribution database 122 . Then, the selection target day narrowing unit 151 calculates the RMSE between the solar radiation distribution at the prediction start time of the prediction target day and the solar radiation distribution for each of the groups #1 to #16 using the paired formula (2). Ask.

FIG. 8 is a diagram for explaining calculation of the RMSE between the distribution of solar radiation on the prediction target day and the distribution of solar radiation for each group. Here, the RMSE in FIG. 8 is obtained by Equation 2 described above. The solar radiation amount distribution at the prediction start time of the prediction target day has data of the solar radiation amount for each predetermined unit area. The solar radiation distribution diagram 61 of FIG. 8 is a diagram showing the solar radiation distribution in the target area at the prediction start time of the prediction target day. Each point 62 arranged in a lattice pattern in the solar radiation amount distribution diagram 61 represents the solar radiation amount for each predetermined unit area. 321 predetermined unit areas are arranged in the column direction and 553 predetermined unit areas are arranged in the row direction in the other small area. Nine predetermined unit regions are arranged in the column direction and nine in the row direction. Here, a case in which 9×9 predetermined unit areas are included will be described as an example.

Each row is numbered so that the number increases by one each time the row goes up with the point at the upper left and lower end of the solar radiation distribution diagram 61 as the origin. Also, the columns are numbered in such a way that the number increases by one each time the column moves to the right. Below, the number of each column is represented by i (i≧1). Also, the number of each row is represented by j (j≧1). That is, each point 62 of the solar radiation amount distribution diagram 61 is expressed as (i, j) with the lower left point on the paper being (1, 1). In FIG. 8, nine predetermined unit areas are arranged in the column direction and nine predetermined unit areas are arranged in the row direction in the target area. That is, the target area includes 9×9 predetermined unit areas.

The solar radiation distribution of group #k (1≤k≤16) also has solar radiation data for each predetermined unit area. A solar radiation distribution diagram 63 in FIG. 8 is a diagram showing the solar radiation distribution for group #k (1≦k≦16). Each point 64 arranged in a grid pattern in the solar radiation amount distribution diagram 63 represents the solar radiation amount for each predetermined unit area. Also in this case, each point 64 of the solar radiation distribution map 63 is expressed as (i, j) with the lower left point as (1, 1) on the paper surface.

Then, the selection target day narrowing unit 151 obtains the RMSE between the solar radiation distribution at the prediction start time of the prediction target day and the solar radiation distribution of each group #1 to #16 using the above-described formula (2). .

Here, the prediction target day (i, j) is the amount of solar radiation at a point 62 represented by (i, j) in the solar radiation distribution diagram 61 . Group #k (i, j) is the amount of insolation at point 64 represented by (i, j) in the insolation amount distribution diagram 63 . And nx*ny is the number of points 62 included in the solar radiation amount distribution map 61 . That is, nx is the maximum value of j and ny is the maximum value of i. Here, nx*ny is 9*9.

That is, the selection target day narrowing-down unit 151 obtains the value obtained by squaring the difference between each point 62 in the solar radiation distribution diagram 61 and each point 64 in the solar radiation distribution diagram 63 for all combinations of (i, j). . Then, the selection target date narrowing unit 151 obtains the square root of the value obtained by dividing the sum of the obtained results by the number of predetermined unit regions included in the target region, and obtains the solar radiation distribution at the prediction start time of the prediction target date and each RMSE, which is the mean square error between the representative solar radiation amounts of groups #1 to #16, is calculated.

Next, the selection target day narrowing unit 151 selects a group having three RMSEs in descending order among the RMSEs calculated using the formula (2) as a solar radiation distribution similar to the prediction start time of the prediction target day. , are selected as similar groups. For example, the selection target date narrowing unit 151 selects groups #6, #7, and #4 as similar groups. Here, although three groups are selected in this embodiment, the number of groups is not limited to three. The selection target date narrowing unit 151 may select a smaller number of groups in order to lighten the subsequent similar date selection processing, and may select a larger number of groups in order to improve the accuracy of the similar date selection processing. It is preferable to select a group of

Next, the selection target date narrowing unit 151 identifies the selection target dates included in the selected similar group using information on the group to which each selection target date determined in advance belongs. Hereinafter, the selection target date specified by the selection target date narrowing unit 151 is referred to as a “comparison target date”.

After that, the selection target date narrowing unit 151 outputs the solar radiation distribution at the prediction start time of the prediction target date and the solar radiation distribution at the prediction start time of the comparison target date to the first difference calculation unit 152 . The selection target date narrowing unit 151 also outputs information on the comparison target date and the prediction target date to the second difference calculation unit 153 . The selection target date narrowing unit 151 also outputs information on the comparison target date and the prediction target date to the relative humidity difference calculation unit 154 . Further, the selection target date narrowing unit 151 outputs the solar radiation distribution at the prediction start time of the prediction target date to the cloud distribution calculation unit 155 .

The first difference calculation unit 152 receives the input of the solar radiation amount distribution at the prediction start time of the prediction target day and the solar radiation amount distribution at the prediction start time of the comparison target day from the selection target day narrowing unit 151 . Then, the first difference calculating unit 152 compares the distribution of solar radiation at the prediction start time of the prediction target day with the distribution of solar radiation at the prediction start time of the prediction target day and the distribution of solar radiation at the prediction start time of the comparison target day. Calculate the RMSE between the solar radiation distribution at the predicted start time of the target day. Specifically, the first difference calculator 152 uses the following formula (3) to calculate the RMSE between the distribution of solar radiation at the prediction start time of the prediction target day and the distribution of solar radiation at the prediction start time of the comparison target day. Calculate

Calculation of the RMSE between the solar radiation distribution at the prediction start time of the prediction target day and the solar radiation distribution at the prediction start time of the comparison target day will be described in detail below with reference to FIG. FIG. 9 is a diagram for explaining the calculation of the RMSE between the solar radiation distribution on the prediction target day and the solar radiation distribution on the comparison target day.

A solar radiation distribution diagram 71 in FIG. 9 represents the solar radiation distribution at the prediction start time of the prediction target day in the target region. Each point 72 arranged in a grid pattern in the solar radiation amount distribution diagram 71 represents the solar radiation amount for each predetermined unit area. Here, each point 72 in the solar radiation amount distribution diagram 71 is expressed as (i, j) with the lower left point as (1, 1) on the paper surface.

A solar radiation distribution diagram 73 in FIG. 9 represents the solar radiation distribution at the prediction start time of the comparison target day in the target region. Each point 74 arranged in a lattice pattern in the solar radiation amount distribution diagram 73 represents the solar radiation amount for each predetermined unit area. Also in this case, each point 74 in the solar radiation distribution diagram 73 is expressed as (i, j) with the lower left point as (1, 1) on the paper surface.

Then, the first difference calculator 152 calculates the RMSE between the solar radiation distribution at the prediction start time of the prediction target day and the solar radiation distribution at the prediction start time of the comparison target day using Equation (3) described above.

Here, the prediction target day (i, j) is the amount of solar radiation at a point 72 represented by (i, j) in the solar radiation distribution diagram 71 . Also, the comparison target day (i, j) is the amount of insolation at point 74 represented by (i, j) in the insolation amount distribution diagram 73 . And nx*ny is the number of points 72 included in the solar radiation amount distribution diagram 71 . Here, nx*ny is 9*9.

That is, the first difference calculator 152 obtains the square of the difference between each point 72 in the solar radiation distribution diagram 71 and each point 74 in the solar radiation distribution diagram 73 for all combinations of (i, j). Then, the first difference calculation unit 152 obtains the square root of the value obtained by dividing the sum of the obtained results by the number of predetermined unit regions included in the target region, and compares the solar radiation distribution at the prediction start time of the prediction target day. Calculate the RMSE between the predicted start time of the day and the solar radiation distribution. The first difference calculator 152 calculates the RMSE of the solar radiation distribution for all comparison days.

After that, the first difference calculation unit 152 outputs the RMSE of the solar radiation distribution for each comparison target date calculated using Equation (3) to the similar date extraction unit 156 . Here, the RMSE between the distribution of solar radiation at the prediction start time of the prediction target day and the solar irradiation at the prediction start time of the comparison target day is the distribution of solar irradiation at the prediction start time of the comparison target day. Represents how much it differs from the solar radiation distribution at the time of day. That is, it can be said that the smaller the RMSE, the more similar the solar radiation distribution at the prediction start time of the comparison target day and the solar irradiation distribution at the prediction start time of the prediction target day.

The second difference calculation unit 153 receives input of information on the prediction target date and the comparison target date from the selection target date narrowing unit 151 . Then, the second difference calculation unit 153 calculates the distribution of solar radiation in the time zone one hour before the prediction start time on the prediction target day and the solar radiation levels in the time zones one hour before and one hour after the prediction start time on the comparison target day. The distribution is obtained from the solar radiation distribution database 122 . Also, the second difference calculator 153 acquires the solar radiation amount distribution for the time zone one hour after the prediction start time of the prediction target day from the PV output prediction result data 124 . Here, the solar radiation amount distribution in the time zone one hour after the prediction start time of the prediction target day is a predicted value by the PV output prediction unit 18 . Here, since the past prediction accuracy is high, the second difference calculation unit 153 obtains the difference distribution of the amount of solar radiation using the distribution of the amount of solar radiation one hour before and one hour after so as to include the past in the calculation. Any other time period may be used as long as it is possible to obtain a change in the amount of solar radiation in the vicinity of the time. For example, when the accuracy of the prediction calculation is high, the second difference calculation unit 153 may use the time period 0 hours before and 2 hours after the prediction start time to calculate the solar radiation amount distribution.

Next, the second difference calculation unit 153 subtracts the solar radiation distribution one hour before the prediction start time on the prediction target day from the solar radiation distribution one hour after the prediction start time on the prediction target day. Generate the solar radiation difference distribution of . In addition, the second difference calculation unit 153 subtracts the solar radiation distribution one hour before the prediction start time on the comparison target day from the solar irradiation distribution one hour after the prediction start time on the comparison target day, Generate a difference distribution of insolation.

The second difference calculation unit 153 uses the difference distribution of solar radiation on the day to be predicted and the difference distribution of solar radiation on each comparison day to calculate the difference between the difference distribution of solar radiation on the day to be predicted and the distribution of difference in radiation on each comparison day. Calculate the RMSE between Specifically, the second difference calculator 153 calculates the RMSE between the solar radiation amount difference distribution of the prediction target day and the solar radiation amount difference distribution of each comparison target day using the following formula (4).

Calculation of the RMSE between the solar radiation amount difference distribution on the prediction target day and the solar radiation amount difference distribution on the comparison target day will be described in detail below with reference to FIG. 10 . FIG. 10 is a diagram for explaining the calculation of the RMSE between the solar radiation amount difference distribution of the prediction target day and the solar radiation amount difference distribution of the comparison target day.

A solar radiation amount difference distribution diagram 81 in FIG. 10 represents the solar radiation amount difference distribution on the prediction target day in the target area. Each point 82 arranged in a grid pattern in the solar radiation amount difference distribution diagram 81 represents the difference in the solar radiation amount between one hour after and one hour before the prediction start time for each predetermined unit area. Here, each point 82 of the solar radiation amount difference distribution diagram 81 is expressed as (i, j) with the lower left point as (1, 1) on the paper surface.

Moreover, the solar radiation amount difference distribution diagram 83 in FIG. 10 represents the solar radiation amount difference distribution on the comparison target day in the target area. Each point 84 arranged in a grid pattern in the solar radiation amount difference distribution diagram 83 represents the difference in the solar radiation amount between the prediction start time and the time period one hour before for each predetermined unit area. Also in this case, each point 84 in the solar radiation amount difference distribution diagram 83 is expressed as (i, j) with the lower left point as (1, 1) on the paper surface.

Then, the second difference calculator 153 obtains the RMSE between the solar radiation amount difference distribution on the prediction target day and the solar radiation amount difference distribution on each comparison target day using the above-described formula (4).

Here, the prediction target day difference (i, j) is the solar radiation amount difference at the point 82 represented by (i, j) in the solar radiation amount difference distribution diagram 81 . Also, the comparison target daily difference (i, j) is the amount of solar radiation at a point 84 represented by (i, j) in the solar radiation amount difference distribution diagram 83 . And nx*ny is the number of points 82 included in the solar radiation amount difference distribution diagram 81 . Here, nx*ny is 9*9.

That is, the second difference calculation unit 153 squares the difference between each point 82 in the solar radiation difference distribution diagram 81 and each point 84 in the solar radiation difference distribution diagram 83 for all (i, j) combinations. Ask. Then, the second difference calculation unit 153 obtains the square root of the value obtained by dividing the sum of the obtained results by the number of predetermined unit areas included in the target area, and calculates the difference distribution of the solar radiation amount on the prediction target day and the solar radiation on the comparison target day. Calculate the RMSE, which is the mean squared error between the volume difference distributions. The second difference calculator 153 calculates the RMSE of the solar radiation amount difference distribution for all comparison days.

After that, the second difference calculation unit 153 outputs the calculated RMSE of the difference distribution of solar radiation for each comparison date to the similar date extraction unit 156 . Here, finding the difference in the amount of solar radiation corresponds to finding the movement of clouds. Therefore, by using the difference distribution of the amount of insolation, it is possible to identify a comparison target day on which cloud movement is similar to that of the forecast target day. In other words, it can be said that the smaller the RMSE, the more similar the motion of the clouds on the day to be compared is to the motion of the clouds on the day to be predicted.

The cloud distribution calculation unit 155 receives the input of the solar radiation amount distribution at the prediction start time of the prediction target day and the solar radiation amount distribution at the prediction start time of the comparison target day from the selection target day narrowing unit 151 . Next, the cloud distribution calculation unit 155 calculates the average value of the solar radiation amount at the prediction start time of the prediction target day using the solar radiation distribution at the prediction start time of the prediction target day. In addition, the cloud distribution calculation unit 155 calculates the average value of the solar radiation amount at the prediction start time of the comparison target day using the solar radiation distribution at the prediction start time of the comparison target day.

Next, the cloud distribution calculator 155 obtains the standard deviation of the solar radiation amount on the prediction target day using the solar radiation distribution at the prediction start time of the prediction target day and the average value of the solar radiation amount on the prediction target day. In addition, the cloud distribution calculation unit 155 obtains the standard deviation of the solar radiation on the prediction target day using the solar radiation distribution at the prediction start time of each comparison target day and the average value of the solar radiation on each comparison target day.

The method of obtaining the standard deviation of the amount of solar radiation is the same for both the prediction target date and the comparison target date. Therefore, with reference to FIG. 11, the calculation of the standard deviation of the solar radiation amount on the prediction target day using the solar radiation distribution at the prediction start time of the prediction target day and the average value of the solar radiation amount on the prediction target day will be described in detail. explain. FIG. 11 is a diagram for explaining the calculation of the standard deviation of the solar radiation amount on the prediction target day using the solar radiation distribution at the prediction start time of the prediction target day and the average value of the solar radiation amount on the prediction target day.

A solar radiation distribution diagram 91 in FIG. 11 represents the solar radiation distribution at the prediction start time of the prediction target day in the target region. Each point 92 arranged in a grid pattern in the solar radiation amount distribution diagram 91 represents the solar radiation amount for each predetermined unit area. Here, each point 92 in the solar radiation amount distribution diagram 91 is expressed as (i, j) with the lower left point as (1, 1) on the paper surface.

The cloud distribution calculation unit 155 sums the values of each point 92 in the solar radiation distribution diagram 91 and divides the total by the number of points 92 included in the solar radiation distribution diagram 91 to obtain the solar radiation amount at the prediction start time of the prediction target day. Find the average value of

Then, the cloud distribution calculator 155 obtains the standard deviation of the amount of solar radiation on the prediction target day using the following formula (5).

Here, the prediction target day (i, j) is the amount of solar radiation at a point 92 represented by (i, j) in the solar radiation distribution diagram 91 . Also, the prediction target day average is the average value of the amount of solar radiation on the prediction target day. And nx*ny is the number of points 92 included in the solar radiation amount distribution diagram 91 . Here, nx*ny is 9*9.

That is, the cloud distribution calculator 155 obtains the square of the difference between each point 92 in the solar radiation distribution diagram 91 and the average value of the solar radiation for all combinations of (i, j). Then, the cloud distribution calculation unit 155 obtains the square root of the value obtained by dividing the sum of the obtained results by the number of predetermined unit areas included in the target area, and calculates the standard deviation of the solar radiation amount on the prediction target day. The second difference calculator 153 similarly calculates the standard deviation of the amount of solar radiation for each comparison target day.

After that, the cloud distribution calculation unit 155 outputs the calculated standard deviation of the amount of solar radiation for the prediction target date and each comparison target date to the similar date extraction unit 156 . Here, the standard deviation of the amount of insolation can be said to be a value summarizing deviations from the average amount of insolation at each point in the target area. That is, it can be said that the larger the standard deviation, the larger the deviation of the amount of solar radiation from the average at many points, and it can be said that many fine clouds are generated. On the other hand, the smaller the standard deviation, the more the amount of solar radiation is the same as the average at most points, and it can be said that there are no clouds. That is, it is possible to determine the existence or non-existence of fine cumulus clouds on sunny days using the standard deviation of the amount of solar radiation.

FIG. 12 is a diagram for explaining a method of determining cumulus clouds during analysis using standard deviations. For example, image 101 is an image when the standard deviation is small. That is, image 101 represents a case where clouds are almost absent. On the other hand, image 102 is an image when the standard deviation is large. That is, the image 102 represents a state in which many fine cumulus clouds are generated. A description will be given assuming that there are states represented by images 103 to 105 as states to be compared with this. In this case, the comparison states represented by images 103-105 have a small standard deviation. And, as can be seen from images 103 to 105, almost no clouds exist. On the other hand, the states represented by images 106 to 108 have large standard deviations. As can be seen from images 106 to 108, many clouds are generated. That is, from the standard deviation, it can be determined that the state of the cumulus clouds in the images 106 to 108 is not similar to the state of the image 101 . Also, from the standard deviation, it can be determined that the state of the cumulus clouds in the images 103 to 105 is not similar to the image 102 .

The relative humidity difference calculation unit 154 receives input of information on the comparison target date and the prediction target date from the selection target date narrowing unit 151 . Then, the relative humidity difference calculator 154 acquires from the relative humidity data 121 the relative humidity RH at the prediction start time at a predetermined observation point on the comparison target date and the prediction target date.

Here, the observation points of the meteorological office that observe the relative humidity in the target area used in this embodiment are the eight points 111 to 118 shown in FIG. FIG. 13 is a diagram showing observation points of the meteorological office in the target area. In this embodiment, the relative humidity difference calculator 154 acquires the relative humidity at the prediction start time of the comparison target date and the prediction target date at the points 111 to 118 .

Next, the relative humidity difference calculator 154 calculates RMSE, which is the root-mean-square error of the relative humidity RH (Relative Humidity) of the comparison target date and the prediction target date, using the following formula (6).

Here, the RH target date time (r) is the relative humidity RH at each of eight points r (1≦r≦8) on the comparison target date. Also, the RH prediction target date (r) is the relative humidity RH at each of eight points r (1≦r≦8) on the prediction target date. Then, the square of the difference between the RH comparison target date (r) and the RH prediction target date (r) is obtained for each of the eight points and summed up. Also, M is the number of observation points. In this example, M=8.

That is, the relative humidity difference calculator 154 obtains the square of the difference between the relative humidity RH on the prediction target date and the relative humidity RH on the comparison target date at each of the points 111 to 118 for all the observation points. Then, the relative humidity difference calculator 154 obtains the square root of the value obtained by dividing the sum of the obtained results by the number of observation points, and calculates the RMSE of the relative humidity RH between the comparison target date and the prediction target date. The relative humidity difference calculator 154 calculates the RMSE of the relative humidity RH for all comparison days.

After that, the relative humidity difference calculation unit 154 outputs the calculated RMSE of the relative humidity between the comparison target date and the prediction target date to the similar date extraction unit 156 . Here, the larger the value of the RMSE of the relative humidity, the more different the humidity state between the comparison target date and the prediction target date. In other words, it can be seen that the relative humidity RH on the comparison target day with a large RMSE is significantly different from the humidity state on the prediction target day.

The similar day extraction unit 156 receives from the first difference calculation unit 152 an input of the RMSE of the solar radiation distribution for each comparison date calculated using Equation (3). Further, the similar day extraction unit 156 receives from the second difference calculation unit 153 the RMSE of the solar radiation amount difference distribution for each comparison target day calculated using Equation (4). The similar day extracting unit 156 also receives from the cloud distribution calculating unit 155 the standard deviation of the amount of solar radiation for the prediction target day and each comparison target day calculated using Equation (5). Further, the similar date extraction unit 156 receives from the relative humidity difference calculation unit 154 the RMSE of the relative humidity between the comparison target date and the prediction target date calculated using Equation (6).

Next, the similar day extraction unit 156 obtains the product of the RMSE of the solar radiation amount distribution and the RMSE of the solar radiation amount difference distribution for each comparison target day. Then, the similar date extracting unit 156 determines the rank in ascending order of the calculation result value. In other words, the similar day extracting unit 156 performs ranking so that comparison days with more similar solar radiation distributions and cloud movement states are ranked higher.

Next, the similar day extraction unit 156 obtains the difference in the standard deviation of the amount of solar radiation between each comparison date and the prediction target date. Then, the similar date extracting unit 156 excludes the comparison dates whose standard deviation of the comparison dates is outside the predetermined range from the standard deviation of the prediction dates from the ranked comparison dates. That is, the similar day extracting unit 156 excludes, from the ranked comparison dates, days on which the state of cumulus clouds is significantly different from the prediction target date.

For example, under sunny conditions, the similar day extraction unit 156 performs the following process of excluding prediction target days. If the standard deviation of the normalized solar radiation amount of the prediction target day is less than 0.05, the similar day extracting unit 156 excludes the comparison target day outside the range of ±0.015. Further, if the standard deviation of the normalized solar radiation amount of the prediction target day is 0.05 or more and less than 0.01, the similar day extraction unit 156 excludes the comparison target day outside the range of ±0.030. Further, if the standard deviation of the normalized solar radiation amount of the prediction target day is 0.0.10 or more, the similar day extracting unit 156 excludes comparison target days outside the range of ±0.050.

Next, the similar date extracting unit 156 excludes comparison dates that are greater than the specified value of the RMES of the relative humidity RH from the ranked comparison dates. That is, the similar day extracting unit 156 excludes days whose humidity conditions are significantly different from the prediction target day from the ranked comparison target days.

For example, in this embodiment, the similar date extracting unit 156 excludes comparison dates with a relative humidity RMSE greater than 10% from the ranked comparison dates. However, it is preferable to change this condition depending on the number of comparison days obtained. For example, if the condition is that the number of samples for the comparison date to be used is 20 or more cases, and the RMSE of the relative humidity is greater than 10% when the comparison date is excluded and the condition is not satisfied, the RMSE It is preferable to relax the conditions for excluding dates from comparison dates by setting the threshold to 20% or 30%. However, even when changing the threshold, it is preferable to determine the maximum value of the threshold, for example, it is preferable to set the maximum value of the threshold to 30%.

Next, the similar date extracting unit 156 moves up the ranking of the remaining comparison dates by the removed comparison dates to determine the ranking. After that, the similar date extraction unit 156 extracts the top three comparison dates as similar dates. The similar date extraction unit 156 then outputs information on the extracted similar dates to the prediction error acquisition unit 16 .

The prediction error acquisition unit 16 receives input of information on similar dates with respect to the prediction target date from the similar date extraction unit 156 of the extraction unit 15 . Then, the prediction error acquisition unit 16 acquires from the prediction error calculation unit 14 the error evaluation result of each prediction model of the group to which the designated similar date belongs. After that, the prediction error acquisition unit 16 outputs the error evaluation result of each prediction model on each acquired similar date to the notification unit 17 .

The PV output prediction unit 18 has in advance first to third prediction models used for PV output prediction. Then, the PV output prediction unit 18 acquires information necessary for prediction from the weather data distribution system 2 . Then, using each prediction model, the PV output prediction unit 18 predicts the PV output in each prediction target time zone of the prediction target date in each case. Then, the PV output prediction unit 18 outputs to the notification unit 17 the prediction result of the PV output when each prediction model is used. The PV output prediction unit 18 stores in the solar radiation distribution database 122 the predicted solar radiation amount distribution calculated for the calculation of the PV output prediction. The predicted solar radiation amount distribution is overwritten with the solar radiation amount distribution of the solar radiation amount distribution estimator 11 at the same time after the prediction.

The notification unit 17 receives an input of the error evaluation result of each prediction model on each similar date from the prediction error acquisition unit 16 . In addition, the notification unit 17 acquires the solar radiation amount distribution of the extracted similar day prediction target start time from the solar radiation amount distribution database 122 . The notification unit 17 also acquires the PV output measurement result on the similar date from the PV output performance data 123 . In addition, the notification unit 17 acquires from the PV output prediction result data 124 the prediction result of the PV output when each prediction model on the similar date is used. Further, the notification unit 17 receives from the PV output prediction unit 18 an input of the prediction result of the PV output on the prediction target day when each prediction model is used. Then, the notification unit 17 uses the distribution of solar radiation at the prediction target start time on the similar day, the measurement result of the PV output on the similar day and each prediction model as the prediction result of the PV output, the error evaluation result of each prediction model on the similar day, and The prediction result of the PV output on the prediction target day when each prediction model is used is transmitted to the terminal device 3 and notified to the user.

An example of information provision by the notification unit 17 when using a web page will be described below with reference to FIGS. 14A, 14B, and 15. FIG. FIG. 14A is a diagram showing the first page of a web page that provides information on a solar radiation amount prediction result. Also, FIG. 14B is a diagram showing the solar radiation amount prediction result of the first page of the web page. Also, FIG. 15 is a diagram showing the second page of the web page that provides information on the PV output prediction result.

As shown in FIG. 14A, a web page 220 displays an image 221 representing the solar radiation distribution at the prediction start time of the prediction target day. Furthermore, the web page 220 displays images 222 to 224 representing the distribution of solar radiation at the predicted start times of each of the three similar days. Furthermore, the web page 220 displays an image 225 representing the average amount of solar radiation for the prediction target day and the prediction result of the amount of solar radiation for each prediction model.

For example, as shown in FIG. 14B, the image 225 also includes observation results, a continuous model based on the average observed value for 10 minutes before the prediction start time, a sunny model assuming fine weather, and a model performed at 10-minute intervals. It is also possible to display the forecasted values based on the forecasted values, the past results of the forecasted values implemented at intervals of 10 minutes, and the forecasted values based on the short-term rainfall forecast. FIG. 14B represents an image 225 showing an example of performing prediction calculations up to 6 hours ahead at 10-minute intervals. A graph 2251 represents the observation results. Also, the solid line portion of the graph 2252 represents the estimated value, and the dashed line portion represents the persistence model based on the estimated value of the previous 10-minute average. A graph 2253 represents a sunny model assuming fine weather. Graph 2254 also represents the predicted values for the 30-minute intervals. A graph 2255 is a predicted value based on the short-term rain forecast by the Japan Meteorological Agency.

For similar days 222-223, images 226-228 representing the amount of solar radiation on the similar days and the prediction results of the amount of solar radiation for each prediction model are displayed. In addition, the web page 220 may display information on groups #1 to #16 similar to the prediction start time period of the prediction start date.

The user can refer to the web page 220 displayed on the terminal device 3 and judge the degree of similarity in the solar radiation distribution between the day to be predicted and each similar day from the images 221 to 224 . In addition, the user refers to the Web page 220 displayed on the terminal device 3 and displays an image 225 of how the average solar radiation amount forecast at the eight meteorological offices in the target area on the forecast target day is. It can be grasped from ∼228. Furthermore, the user can grasp from the images 225 to 228 how much the prediction result of the solar radiation amount of each prediction model deviated from the actual measurement result in the past prediction on each similar day.

Further, when the user selects the display of the second page, the terminal device 3 displays the web page 230 shown in FIG. The web page 230 displays images 231 to 233 showing evaluation results for each similar day. In addition, on the web page 230, images 234 to 236 showing the measurement results and prediction results of the PV output on each similar day are displayed.

By referring to the images 231 to 233, the user can determine which prediction model is most likely to make the most appropriate prediction for a date and time similar to the prediction time period of the prediction target date. In addition, by referring to the images 234 to 236, the user can confirm which prediction model has actually predicted the PV output with what degree of accuracy in a date and time similar to the prediction time period of the prediction target day. can be done. That is, by referring to the web pages 220 and 230, the user can select the prediction model that is considered to have the highest accuracy in the prediction time period of the prediction target date.

Next, with reference to FIG. 16, the flow of notification processing of the prediction error of each prediction model by the PV output prediction device 1 according to the present embodiment will be described. FIG. 16 is a flowchart of processing for notifying the prediction error of each prediction model by the PV output prediction device according to the first embodiment.

The classification unit 13 acquires the solar radiation distribution in the predicted time period of the past day from the solar radiation distribution database 122 . Then, the classification unit 13 performs hierarchical clustering to classify past days into 16 groups 1 to #16 (step S1).

The prediction error calculation unit 14 acquires the measurement result of the PV output in the prediction target time zone of the past day from the PV output performance data 123, and also obtains the PV output prediction result of each prediction model in the prediction target time zone of the past day. is obtained from the PV output prediction result data 124 . Next, the prediction error calculation unit 14 evaluates the error of each prediction model for each group #1 to #16 generated by the classification unit 13, and calculates the prediction error for each prediction model in the prediction target time zone of the past day. (Step S2).

The extracting unit 15 extracts a comparison target date from the selection target dates included in the past dates. Then, the extracting unit 15 executes a similar date extraction process using the prediction start time of the prediction target date and the comparison target date, the solar radiation distribution before and after one hour, and the relative humidity of the prediction target date and the comparison target date. (Step S3).

The prediction error acquisition unit 16 acquires the prediction error of the similar date extracted by the extraction unit 15 from the prediction error calculation unit 14 (step S4).

The notification unit 17 outputs the prediction error of the similar date acquired by the prediction error acquisition unit 16 to the terminal device 3 to notify the user (step S5). At this time, the notification unit 17 may output a prediction result or the like when using each prediction model of PV output.

Next, with reference to FIG. 17, the flow of extraction processing of similar dates by the extraction unit 15 will be described. FIG. 17 is a flowchart of a similar date extraction process.

The selection target date narrowing unit 151 acquires the classification result of the solar radiation distribution and the solar radiation distribution of each group from the classification unit 13 . Further, the selection target date narrowing unit 151 acquires the solar radiation distribution at the prediction start time of the selection target date from the solar radiation distribution database 122 . Furthermore, the selection target date narrowing unit 151 selects 45 days before and after the same date as the prediction target date from the past days as selection target dates. Next, the selection target date narrowing unit 151 specifies a group to which the solar radiation distribution of the prediction start time of each selection target date belongs from the classification results obtained from the classification unit 13 (step S101).

Next, the selection target date narrowing unit 151 acquires the solar radiation distribution at the prediction start time of the prediction target date from the solar radiation distribution database 122 . Then, the selection target day narrowing unit 151 calculates the RMSE between the solar radiation distribution at the prediction start time of the prediction target day and the solar radiation distribution for each group (step S102).

Next, the selection target date narrowing-down unit 151 sets the top three groups with the smallest RMSE of the solar radiation amount distribution as similar groups. Then, the selection target date narrowing unit 151 selects the selection target date belonging to the similar group as the comparison target date, and narrows down the comparison target date (step S103).

The first difference calculation unit 152 acquires the solar radiation amount distributions of the prediction start times of the prediction target date and the comparison target date from the selection target date narrowing unit 151 . Then, the first difference calculator 152 calculates the RMSE of the solar radiation amount distribution between the prediction start time of the prediction target date and the comparison target date using Expression (3) (step S104).

The second difference calculation unit 153 acquires information on the prediction target date and the comparison target date from the selection target date narrowing unit 151 . Then, the second difference calculator 153 acquires from the solar radiation distribution database 122 the solar radiation distribution for the time period around one hour before and after the prediction start time of the prediction target day and the comparison target day. Here, the solar radiation amount distribution one hour after the prediction start time of the prediction target day is a predicted value. Next, the second difference calculation unit 153 calculates the difference between the solar radiation amount distribution in the time period one hour before the prediction start time and the solar radiation amount distribution in the time period one hour after the prediction start time of each of the prediction start time and the comparison target day. Then, the solar radiation amount difference distribution is obtained. Next, the second difference calculation unit 153 calculates the RMSE between the solar radiation amount difference distribution on the prediction target day and the solar radiation amount difference distribution on each target day using Expression (4), and calculates the movement of the cloud area. A speed difference is calculated (step S105).

The cloud distribution calculation unit 155 acquires the solar radiation amount distribution of the prediction start time of the prediction target date and the comparison target date from the selection target date narrowing unit 151 . Next, the cloud distribution calculation unit 155 calculates the average value of the solar radiation amount at the prediction start time of the prediction target day and the comparison target day from the solar radiation distribution at the prediction start time of the prediction target day and the comparison target day. Next, the cloud distribution calculation unit 155 uses the solar radiation distribution and the average value of the solar radiation at the prediction start time of the prediction target day and the comparison target day in Equation (5), Each cloud distribution is obtained by obtaining the standard deviation of the amount of solar radiation (step S106).

The relative humidity difference calculation unit 154 acquires information on the prediction target date and the comparison target date from the selection target date narrowing unit 151 . Next, the relative humidity difference calculator 154 acquires from the relative humidity data 121 the relative humidity RH at each of the predetermined observation points at the prediction start times of the prediction target date and the comparison target date. Then, the relative humidity difference calculating unit 154 calculates the RMSE between the relative humidity RH at the measurement start time of the prediction target day and the relative humidity RH at the measurement start time of the comparison target day using Expression (6) (step S107).

The similar day extraction unit 156 receives from the first difference calculation unit 152 the RMSE of the solar radiation distribution for each comparison date. Further, the similar day extracting unit 156 receives an input of the RMSE of the difference distribution of solar radiation for each comparison target day from the second difference calculating unit 153 . The similar day extracting unit 156 also receives from the cloud distribution calculating unit 155 an input of the standard deviation of the amount of solar radiation for the prediction target day and each comparison target day. Furthermore, the similar date extraction unit 156 receives an input of the RMSE of the relative humidity between the comparison target date and the prediction target date from the relative humidity difference calculation unit 154 . Then, the similar day extraction unit 156 obtains the product of the RMSE of the solar radiation distribution and the RMSE of the solar radiation distribution for each comparison target day. Then, the similar date extracting unit 156 determines the rank in ascending order of the calculation result value. That is, the similar day extracting unit 156 performs ranking so that comparison days with more similar solar radiation distributions and cloud movement states are ranked higher (step S108).

Next, the similar day extraction unit 156 obtains the difference in the standard deviation of the amount of solar radiation between each comparison date and the prediction target date. Then, the similar date extracting unit 156 determines, as an exclusion date, a comparison target date having a large value difference with respect to the standard deviation of the calculated prediction target date. In addition, the similar date extraction unit 156 also determines a comparison target date with a large RMSE value of the relative humidity as an exclusion date. Then, the similar date extraction unit 156 excludes the excluded dates from the ranked comparison dates (step S109). That is, the similar day extracting unit 156 excludes days on which the cumulus state is significantly different from the prediction target day and days on which the humidity state is significantly different from the prediction target day, from the ranked comparison dates.

The similar date extracting unit 156 resets the order of the dates to be compared, excluding the excluded dates, in descending order, and determines the order. Then, the similar date extracting unit 156 determines the three comparison dates with the highest rankings as dates similar to the prediction target date (step S110).

After that, the similar date extraction unit 156 notifies the prediction error acquisition unit 16 of the determined similar date (step S111).

As described above, the PV output prediction device according to the present embodiment classifies the past days into groups having common characteristics from the solar radiation distribution, and calculates the prediction error of each prediction model for each group. Then, a similar day having a solar radiation environment similar to the day to be predicted is identified, and the user is notified of the prediction error of each prediction model of the group to which the similar day belongs. Thereby, the user can determine the prediction model to be used for predicting the PV output based on the prediction error obtained by rational calculation, instead of relying on experience. That is, the PV output prediction device according to the present embodiment can contribute to the improvement of the prediction accuracy of the PV output.

In addition, the PV output prediction apparatus according to the present embodiment identifies similar days using the solar radiation environment, the solar radiation amount distribution, the cloud movement state, the cloud distribution, and the relative humidity RH. In this way, by specifying a similar date covering many conditions that affect the PV output, it is possible to improve the accuracy of specifying a similar date whose PV output is similar to the prediction target date. Output prediction accuracy can be improved.

Furthermore, in identifying similar days, the PV output prediction device according to the present embodiment extracts selected days in the same season as the prediction target day from past days, and further extracts the solar radiation distribution from the selected days. is narrowed down to comparison dates that belong to a group similar to the forecast date. As a result, it is possible to reduce the number of objects to be compared when identifying similar dates, reduce the processing load, and identify similar dates at high speed.

Furthermore, the PV output prediction device according to the present embodiment provides the user with a large amount of information such as the distribution of solar radiation on a prediction target day and similar days, the result of solar radiation prediction, the error evaluation result for each prediction model, and the PV output prediction result. provide to This allows the user to evaluate prediction models from various angles and select a more appropriate prediction model. In other words, the PV output prediction device according to the present embodiment can contribute to the improvement of the PV output prediction accuracy in terms of the types of information to be provided.

Here, in the present embodiment, the PV output prediction device 1 having the PV output prediction unit 18 has been described. The PV output prediction support device may have the remaining functions such as acquisition. Even in this case, the PV output prediction support device can contribute to the improvement of the prediction accuracy of the PV output by providing the user with the prediction error of the similar date. That is, the PV output prediction device 1 corresponds to an example of a "PV prediction support device".

Furthermore, in this embodiment, the PV output prediction device 1 has the solar radiation distribution estimating unit 11, but the function of the solar radiation distribution estimating unit 11 is given to another device, and the solar radiation distribution information obtained by the device is may be acquired and used by the PV output prediction device 1 .

FIG. 18 is a block diagram of a PV output prediction device according to the second embodiment. The PV output prediction device 1 according to the present embodiment has a usage prediction model determination section 19 in addition to each section of the first embodiment. The PV output prediction device 1 according to this embodiment differs from the first embodiment in that it automatically determines a prediction model to be used for PV output prediction. In the following description, unless otherwise specified, the parts having the same reference numerals as those in FIG. 2 have the same functions, and description thereof will be omitted.

The used prediction model determining unit 19 acquires the similar date and the error evaluation result of the prediction model on each similar date from the prediction error acquiring unit 16 . Next, the use prediction model determining unit 19 uses the result of the error evaluation to determine the order of each prediction model for each prediction time period of each similar day. That is, the predictive model to be used determination unit 19 ranks the predictive model with the highest error evaluation result in a certain time slot.

Next, the use prediction model determining unit 19 gives scores to the order of each prediction model for each similar date, which is the top three comparison dates. A higher score is given as the result of the error evaluation is higher. In addition, weighting is given to the higher compared dates, and the total score of each prediction model for the higher three compared dates is calculated. Then, the predictive model to be used determination unit 19 determines the predictive model with the highest calculated specific total value as the predictive model to be used. After that, the usage prediction model determining unit 19 notifies the notification unit 17 of the determined usage prediction model.

The notification unit 17 receives the usage prediction model notification from the usage prediction model determination unit 19 . In addition, the notification unit 17 acquires from the PV output prediction result data 124 the prediction result of the PV output at the prediction start time of the prediction target day, which is performed using the designated usage prediction model. Then, the notification unit 17 transmits to the terminal device 3 the prediction result of the PV output at the prediction start time of the prediction target day, which is performed using the usage prediction model, and notifies the user.

Here, in this embodiment, the PV output prediction device 1 notifies the user of the prediction result using the determined usage prediction model. may notify you.

As described above, the PV output prediction device according to the present embodiment determines a prediction model to be used for prediction from the error evaluation result, and provides the user with the PV output prediction result using the determined prediction model. do. As a result, the PV output prediction device according to the present embodiment can obtain the PV output prediction result using the rationally selected prediction model, and can improve the PV output prediction accuracy. In addition, the user can acquire the prediction result of the PV output using the rationally selected and considered appropriate prediction model, and can appropriately control the PV output.

Next, Example 3 will be described. The PV output prediction device according to this embodiment differs from that of the first embodiment in that further information is provided for selecting a prediction model. The PV output prediction device according to this embodiment is also represented by the block diagram of FIG. In the following description, unless otherwise specified, the parts having the same reference numerals as those in FIG. 2 have the same functions, and description thereof will be omitted.

The notification unit 17 acquires from the solar radiation distribution database 122 the solar radiation distribution for each time period included in the three hours before and after the prediction start time of the prediction target day and the similar day. As illustrated in FIG. 19 , the notification unit 17 generates a web page displaying an image 303 representing the solar radiation amount distribution at the prediction start time of the prediction target date and the similar date. FIG. 19 is a diagram illustrating an example of a web page provided by the PV output prediction device according to the third embodiment;

In addition, the notification unit 17 causes the generated web page to display a time selection button 301 for displaying images of predetermined times before and after the predicted start time. Furthermore, the notification unit 17 displays a slide bar 302 for continuously changing the time on the Web page. Then, the notification unit 17 causes the terminal device 3 to display the generated web page and provides information to the user.

Then, when the user selects one of the time selection buttons 301 using the input device of the terminal device 3, the notification unit 17 displays an image of the solar radiation amount distribution map in the time period corresponding to the selected time selection button 301. 303 is displayed. When the user uses the input device of the terminal device 3 to select a continuous time from a specific time with the slide bar 302, the notification unit 17 selects a continuous time from the solar radiation amount distribution of the time zone according to the selected time. The image 303 is switched to display the solar radiation amount distribution of the time period.

Further, when the notification unit 17 receives an input of an instruction to provide more detailed information for selecting a prediction model using the input device of the terminal device 3, the notification unit 17 receives a similar The extraction unit 15 is requested to extract the date. After that, the notification unit 17 acquires from the extraction unit 15 the information on the similar dates in each time slot of the predetermined period and the information on the group to which each similar date belongs. Then, as shown in FIG. 20, the notification unit 17 displays a Web page displaying similar date information 311 representing information on similar dates in each time zone of a predetermined period and group information 312 representing information on the group to which each similar date belongs. is displayed on the terminal device 3 and provided to the user. FIG. 20 is a diagram showing an example of another web page provided by the PV output prediction device according to the third embodiment.

By using the time selection button 301 and the slide bar 302, the user can grasp the change in the distribution of the amount of solar radiation around the predicted time period, and can confirm the movement of clouds from the change in the distribution of the amount of solar radiation. As a result, the user can select a more appropriate prediction model by predicting changes in the weather from the movement of clouds around the prediction time period and using it for selection of a prediction model.

In addition, by obtaining information on similar days in each time period of a predetermined period including the predicted time period and the group to which those similar days belong, the user can obtain the appearance rate of similar days near the predicted time period and the appearance of groups. The rate can be grasped, the similarity rate for each similar day and the prediction target day of the group can be confirmed, and a more appropriate prediction model can be selected.

As described above, the PV output prediction device according to the present embodiment notifies the user of the movement of clouds in the vicinity of the prediction time period, similar days, and appearance of groups. As a result, the user can select a more appropriate prediction model, so that the PV output prediction device according to the present embodiment can contribute to improving the prediction accuracy of the PV output.

(Hardware configuration)
Next, with reference to FIG. 21, the hardware configuration of the PV output prediction device 1 described in each of the above embodiments will be described. FIG. 21 is a hardware configuration diagram of a PV output prediction device.

The PV output prediction device 1 has a CPU (Central Processing Unit) 901, a memory 902, a hard disk 903, and a network interface 904, as shown in FIG. A CPU 901 is connected to a memory 902, a hard disk 903, and a network interface 904 via a bus.

A hard disk 903 is an auxiliary storage device and implements the function of the storage unit 12 . The hard disk 903 includes the solar radiation distribution estimation unit 11, the classification unit 13, the prediction error calculation unit 14, the extraction unit 15, the prediction error acquisition unit 16, the notification unit 17, and the PV output prediction shown in FIGS. Stores various programs including programs for realizing the functions of the unit 18 and the usage prediction model determination unit 19 . Also, the network interface 904 is an interface for communication with the weather data delivery system 2 and the terminal device 3 .

The CPU 901 reads various programs from the hard disk 903, develops them on the memory 902, and executes them, thereby obtaining the solar radiation distribution estimation unit 11, the classification unit 13, the prediction error calculation unit 14, and the extraction The functions of the unit 15, the prediction error acquisition unit 16, the notification unit 17, the PV output prediction unit 18, and the usage prediction model determination unit 19 are realized.

Note that the solar radiation distribution estimation unit 11, the classification unit 13, the prediction error calculation unit 14, the extraction unit 15, the prediction error acquisition unit 16, the notification unit 17, the PV output prediction unit 18, and the usage prediction model determination unit of each embodiment described above The program for realizing each function of 19 does not necessarily have to be stored in the hard disk 903 from the beginning.

For example, a “portable physical medium” such as a flexible disk (FD), a CD-ROM (Compact Disk-Read Only Memory), a DVD, a magneto-optical disk, an IC (Integrated Circuit) card inserted into the PV output prediction device 1 store the program in the Then, the PV output prediction device 1 may read and execute the program from these.

Furthermore, the program is stored in "another computer (or server)" or the like connected to the PV output prediction device 1 via a public line, the Internet, a LAN (Local Area Network), a WAN (Wide Area Network), or the like. back. Then, the PV output prediction device 1 may read and execute the program from these.

1 PV output prediction device 2 weather data distribution system 3 terminal device 11 solar radiation distribution estimation unit 12 storage unit 13 classification unit 14 prediction error calculation unit 15 extraction unit 16 prediction error acquisition unit 17 notification unit 18 PV output prediction unit 19 usage prediction model Determination unit 121 Relative humidity data 122 Solar radiation distribution database 123 PV output performance data 124 PV output prediction result data 151 Selection target day narrowing unit 152 First difference calculation unit 153 Second difference calculation unit 154 Relative humidity difference calculation unit 155 Cloud distribution Calculation unit 156 Similar date extraction unit

Claims (10)

A storage unit for storing information on past-day solar radiation distribution in a target area, past-day photovoltaic power output measurement results, and past-day prediction results for each of a plurality of prediction models used for predicting photovoltaic power output. When,
an extraction unit that acquires information on the distribution of solar radiation for a prediction target day, and extracts similar days having the distribution of solar radiation similar to the prediction target day from selected days included in a predetermined period of the past days; ,
a prediction error acquisition unit that acquires a prediction error for each prediction model on the similar date extracted by the extraction unit based on the measurement result and the prediction result for each prediction model;
and a notification unit that notifies each of the prediction errors acquired by the prediction error acquisition unit as a prediction error of each of the prediction models for predicting the solar power generation output on the prediction target day. support equipment. a classification unit that classifies the past days into a plurality of groups based on the solar radiation distribution;
a prediction error calculation unit that acquires the prediction result for each prediction model for each group and calculates the prediction error for each prediction model for each group;
2. The prediction error acquisition unit acquires the prediction error for each prediction model calculated by the prediction error calculation unit for the group to which the similar date extracted by the extraction unit belongs. The PV output prediction support device according to 1. 3. The PV output prediction support device according to claim 2, wherein the prediction error calculator obtains the prediction error using an error evaluation function based on the measurement result and the prediction result of the past day. The extracting unit identifies the group to which each of the selected target days belongs based on the solar radiation distribution of the selected target day, and determines the prediction target day for each of the groups based on the solar radiation distribution of the prediction target day. 4. The PV output prediction support device according to claim 2 or 3, wherein a similar date is obtained from the selection target dates belonging to a predetermined number of the groups having a small error. The extraction unit obtains a predetermined weather state of the prediction target day and each of the selected target days based on the solar radiation distribution of the prediction target day and each of the selected target days, The PV output prediction support device according to any one of claims 1 to 4, wherein the similar days are extracted based on the solar radiation distribution and the predetermined weather conditions. The extractor is
a first difference calculation unit that calculates a first difference between the distribution of solar radiation on the prediction target day and the distribution of solar radiation on each of the selected days;
Obtaining the moving speed of the cloud area on the prediction target day from the solar radiation distribution on the prediction target day, obtaining the moving speed of the cloud area on each of the selected target days from the solar radiation distribution on each of the selected target days, and obtaining the cloud area moving speed on each of the selected target days a second difference calculation unit that calculates a second difference between the moving speed of the cloud area and the moving speed of the cloud area on each of the selection target days;
obtaining a cloud distribution for each of the selected days based on the solar radiation distribution for each of the selected days; a cloud distribution difference acquisition unit that calculates a difference in the cloud distribution from each of the selection target days;
a humidity difference acquisition unit that acquires the relative humidity of each of the selection target days and the prediction target days, and calculates the difference in the relative humidity between each of the selection target days and the prediction target days;
Based on the first difference and the second difference, selected from the selection target days excluding the day on which the difference in cloud distribution is greater than a first predetermined value and the day on which the difference in relative humidity is greater than a second predetermined value The PV output prediction support device according to any one of claims 1 to 5, further comprising: a similar date extraction unit that extracts the similar date. further comprising a use prediction model determination unit that determines, from among the prediction models, a use model to be used for predicting the solar power generation output on the prediction target day based on each of the prediction errors acquired by the prediction error acquisition unit. The PV output prediction support device according to any one of claims 1 to 6, characterized by: a storage unit for storing information on past-day solar radiation distribution in a target area, past-day photovoltaic output measurement results, and past-day prediction results for each of a plurality of prediction models used to predict photovoltaic power output; ,
an extraction unit that acquires information on the distribution of solar radiation on a prediction target day and extracts similar days having a distribution of solar radiation that is similar to the prediction target day from selected days included in a predetermined period of the past days;
a prediction error acquisition unit that obtains a prediction error for each prediction model on the similar date extracted by the extraction unit based on the measurement result and the prediction result for each prediction model;
a decision unit that decides, from among the prediction models, a model to be used for predicting the solar power generation output on the prediction target day based on the prediction error obtained by the prediction error obtaining unit;
A PV output prediction device, comprising: a prediction unit that predicts a photovoltaic power generation output using the usage model determined by the determination unit. Acquiring information on the distribution of solar radiation for the past day in the target area, measurement results of the photovoltaic power generation output for the past day, and prediction results for the past day for each of a plurality of prediction models used for predicting the photovoltaic power output,
obtaining information on the distribution of solar radiation for a prediction target day, and extracting similar days having a distribution of solar radiation similar to the prediction target day from selected target days included in a predetermined period of the past days;
Based on the measurement result of the past day and the prediction result of each prediction model, obtain a prediction error of each prediction model on the extracted similar date,
A PV output prediction support method, wherein each of the obtained prediction errors is notified as a prediction error of each of the prediction models for predicting the photovoltaic power generation output on the prediction target day. Acquiring information on the distribution of solar radiation for the past day in the target area, measurement results of the photovoltaic power generation output for the past day, and prediction results for the past day for each of a plurality of prediction models used for predicting the photovoltaic power output,
Acquiring information on the distribution of solar radiation for a prediction target day, extracting similar days having similar distributions of solar radiation among the past days,
Obtaining a prediction error for each prediction model on the selected similar date based on the measurement result on the past day and the prediction result for each prediction model;
A PV output prediction support program characterized by notifying each of the obtained prediction errors as a prediction error of each of the prediction models for predicting the photovoltaic power generation output on the prediction target day.

Images