JP7305462B2 - Solar radiation forecasting device, solar radiation forecasting method and solar radiation forecasting program - Google Patents

Description

The present invention relates to a solar radiation prediction device, a solar radiation prediction method, and a solar radiation prediction program.

BACKGROUND ART Photovoltaic (PV) power generation is attracting attention as clean energy for the realization of a low-carbon society and the efficient use of energy. For this reason, it is expected that the introduction of photovoltaic power generation systems will expand in order to reduce the burden on the global environment.

In general, a photovoltaic power generation system directly outputs the amount of power generation that fluctuates depending on weather conditions to an interconnected power system. Therefore, when the scale of the photovoltaic power generation system increases, there is a risk that the power system will become unstable due to fluctuations in the generated power. Below, the output of a photovoltaic power generation system is called PV output. Conventionally, the PV output is predicted in advance, demand control is performed using the predicted value, and stabilization of the entire power system to which the photovoltaic power generation system is interconnected has been performed.

Therefore, in power system operation, it is important to predict fluctuations in photovoltaic power generation output in advance. Therefore, there is a need for a method of predicting with high accuracy the temporal change in solar radiation, which is a factor in fluctuations in photovoltaic power generation output. Numerical weather forecasting is currently frequently used as this forecasting method. Numerical weather forecasting is a method of simulating various weather phenomena by solving physical equations. However, it may be difficult to obtain the prediction accuracy required for power system operation only with this prediction method, and further improvement in accuracy is desired.

The following techniques have been proposed as techniques for further improving the accuracy of numerical weather forecasts. For example, regression that minimizes the sum of the prediction error between the predicted values calculated by applying the actual values of weather conditions to the multiple explanatory variables of the regression formula and the actual values, and the penalty term proportional to the regression coefficient of the regression formula. There is a conventional technique of calculating coefficients and making predictions using a regression equation to which the calculated regression coefficients are applied. In addition, there is a conventional technique for estimating a confidence interval in a method of predicting the amount of solar radiation using numerical prediction data, taking into consideration the tendency of error distribution with respect to predicted values for solar radiation. These are techniques for statistically generating a more accurate prediction formula using past data on actually measured solar radiation amounts.

Geostationary meteorological satellites can also be used to predict solar radiation over a wide area. The data measured by geostationary meteorological satellites are two-dimensional surface data such as photographs of the earth taken from satellites called satellite images. Satellite images are wide-area and high-resolution information. For example, satellite images of Himawari-8, a geostationary meteorological satellite, include information obtained by measuring an area including all of Japan with a spatial resolution of about 1 km. Moreover, satellite images provide meteorological information that is difficult to obtain with the limited number of observation devices installed on the ground.

A method for estimating the current state of solar radiation is a major technology related to solar radiation prediction using satellite images. The current solar radiation estimation method is a method of estimating the solar radiation at that time from a satellite image at that time. In the current solar radiation estimation method, solar radiation is estimated by estimating the distribution of clouds from the reflected light of clouds obtained from satellite images.

For example, in a method of estimating the current solar radiation situation using Himawari-8 satellite image data, an estimated solar radiation value with a spatial resolution of about 1 km can be obtained. Estimating the amount of solar radiation with a spatial resolution of 1 km corresponds to calculating estimated solar radiation values at representative points set at intervals of 1 km. Since the points where solar radiation is observed on the ground are limited, this method for estimating current solar radiation conditions is useful as a means of obtaining solar radiation information over a wide area.

JP 2016-224566 A JP-A-04-167046 WO2004/027676

Naoto Ishibashi, Tatsuya Iizaka, Ryoko Ohira, Yosuke Nakanishi, "Prediction of Solar Radiation Amount Using Partial Least Squares Method and Estimation of Its Confidence Interval," The Institute of Electrical Engineers of Japan Transactions B Atsushi Hashimoto, Akira Usami, "Development of Solar Radiation Estimation and Prediction System Using Himawari-8," Report of Central Research Institute of Electric Power Industry N16001, 2017

However, the conventional technology that statistically generates a prediction formula using past data of actually measured solar radiation is limited to observation points for the actual solar radiation used for prediction. Not suitable for prediction. On the other hand, the solar radiation prediction using the current solar radiation estimation method does not have sufficient prediction accuracy for the amount of solar radiation.

The technology disclosed herein has been made in view of the above, and aims to provide a solar radiation prediction device, a solar radiation prediction method, and a solar radiation prediction program capable of highly accurate short-term future prediction of solar radiation.

In one aspect of the solar radiation prediction device, the solar radiation prediction method, and the solar radiation prediction program disclosed in the present application, the solar radiation estimated value calculation unit calculates the solar radiation estimated value from the satellite image. The prediction formula generation unit is based on the solar radiation estimated value of the past prediction execution time calculated based on the satellite image , the first solar radiation predicted value calculated by numerical weather forecast based on the weather information , and the weather information Using the second solar radiation prediction value calculated by predicting cloud movement as an input variable, a prediction formula is generated based on the solar radiation estimation at the past prediction target time calculated by the first solar radiation estimation value calculation unit. The prediction execution unit performs solar radiation prediction at the prediction target time using the prediction formula generated by the prediction formula generation unit .

In one aspect, the present invention can improve the accuracy of short-term future prediction of the amount of solar radiation.

FIG. 1 is a schematic configuration diagram of a solar radiation prediction system. FIG. 2 is a block diagram of a solar radiation prediction device. FIG. 3 is an example of a satellite image with a center wavelength of 0.64 μm. FIG. 4 is an example of a satellite image with a center wavelength of 10.4 μm. FIG. 5 is an example of a satellite image with a center wavelength of 12.3 μm. FIG. 6 is a diagram showing an overview of the solar radiation amount prediction processing by the solar radiation prediction device according to the first embodiment. FIG. 7 is a flowchart of processing for generating a solar radiation amount prediction formula by the solar radiation prediction device according to the first embodiment. FIG. 8 is a flowchart of solar radiation amount prediction processing by the solar radiation prediction device according to the first embodiment. FIG. 9 is a diagram showing prediction target points when the resolution is reduced. FIG. 10 is a conceptual diagram of a technique for suppressing overlearning using intermediate variables according to the third embodiment. FIG. 11 is a diagram for explaining the accuracy of solar radiation prediction by the solar radiation prediction device.

Hereinafter, embodiments of the solar radiation prediction device, the solar radiation prediction method, and the solar radiation prediction program disclosed in the present application will be described in detail based on the drawings. The solar radiation prediction device, the solar radiation prediction method, and the solar radiation prediction program disclosed in the present application are not limited to the following embodiments.

FIG. 1 is a schematic configuration diagram of a solar radiation prediction system. As shown in FIG. 1 , the solar radiation prediction system 10 has a solar radiation prediction device 1 , a geostationary meteorological satellite 2 , a weather information distribution system 3 and terminal devices 4 . In the following description, the amount of solar radiation (unit: J/m ² or the like) will be used as an example of solar radiation, but the same applies to solar radiation intensity (unit: W/m ² or the like).

The geostationary meteorological satellite 2 is a meteorological satellite orbiting at the same speed as the rotation period of the earth. In this embodiment, the geostationary meteorological satellite 2 includes the whole of Japan in its observation area. The geostationary meteorological satellite 2 observes the entire earth within the range visible by the visible and infrared radiometer. The geostationary meteorological satellite 2 acquires, for example, visible image data and infrared image data of the entire visible range of the earth. Albedo (reflectance) information reflected from the ground surface and clouds can be obtained from the visible image data. Visible images are used to determine cloud thickness and classify stratiform and convective clouds. Also, temperature information obtained from the amount of radiation emitted from the ground surface and clouds can be obtained from the infrared image data. Infrared images are used to identify cloud top heights and clouds.

The geostationary meteorological satellite 2 can communicate with the solar radiation prediction device 1 . The geostationary meteorological satellite 2 transmits the acquired infrared image data and visible image data to the solar radiation prediction device 1 . However, the solar radiation prediction device 1 does not have to directly acquire data from the geostationary meteorological satellite 2. For example, the data transmitted from the geostationary meteorological satellite 2 is accumulated in the weather information distribution system 3 or the like, and the solar radiation prediction device 1 , the accumulated data may be obtained.

The weather information distribution system 3 accumulates GPV (Grid Point Value) data including weather data such as temperature, atmospheric pressure, wind speed, humidity, and amount of solar radiation observed in various places. Here, the GPV data is grid point data of weather elements created as forecast materials by the Japan Meteorological Agency. A weather information delivery system 3 is connected to the solar radiation prediction device 1 . The weather information delivery system 3 provides the solar radiation prediction device 1 with specified weather data among the GVP data.

The terminal device 4 is a device used by a user who uses the prediction results of the solar radiation prediction device 1 . A terminal device 4 is connected to the solar radiation prediction device 1 . The terminal device 4 receives the prediction result of solar radiation prediction from the solar radiation prediction device 1 and provides it to the user.

The solar radiation prediction device 1 uses the infrared image data and visible image data generated by the geostationary meteorological satellites 2 and the weather data acquired from the weather information distribution system 3 to perform solar radiation prediction for a date and time designated as a prediction target. . Then, the solar radiation prediction device 1 outputs the prediction result of the solar radiation prediction to the terminal device 4 . Details of the solar radiation prediction device 1 will be described below with reference to FIG. 2 . FIG. 2 is a block diagram of a solar radiation prediction device.

The solar radiation prediction device 1 has, as shown in FIG.

The solar radiation estimated value calculator 101 receives satellite image data from the geostationary meteorological satellite 2 . Then, the solar radiation estimated value calculation unit 101 assumes that extra-atmospheric solar radiation is attenuated in the atmosphere and reaches the ground surface, and obtains the attenuation from the satellite image with the central wavelength of 0.64 μm illustrated in FIG. Calculate the estimated amount of solar radiation at the time when the satellite image data was captured. FIG. 3 is an example of a satellite image with a center wavelength of 0.64 μm. The vertical axis in FIG. 3 represents latitude, and the horizontal axis represents longitude. In addition, FIG. 3 represents the amount of reflected light by shading indicated by a band arranged on the right side of the paper. Specifically, the solar radiation estimated value calculation unit 101 calculates the solar radiation amount estimated value using the following formula (1).

Here, the solar radiation estimated value calculation unit 101 obtains the reflectance of the estimation target point obtained from the satellite image with the center wavelength of 0.64 μm as the cloud albedo. In addition, the solar radiation estimated value calculation unit 101 obtains the minimum value of the cloud albedo at the estimation target time at the estimation target point for the most recent 20 days as the ground surface albedo. The last term in formula (1), (1-cloud albedo)/(1-ground surface albedo), represents the attenuation ratio of extra-atmospheric solar radiation passing through the atmosphere. It approaches 1 when there are few clouds.

In addition, the solar radiation estimated value calculation unit 101 uses a correction coefficient obtained from past data for more accurate estimation. For example, the solar radiation estimated value calculation unit 101 acquires past solar radiation estimated values calculated by itself from the data storage unit 102 . In addition, the solar radiation estimated value calculation unit 101 acquires the solar radiation observation value of the weather office from the data storage unit 102 . Then, the estimated solar radiation value calculation unit 101 determines a correction coefficient so as to minimize the difference between the estimated solar radiation amount and the observed solar radiation value for the last 30 days. Furthermore, the solar radiation estimated value calculation unit 101 calculates the following by the brightness temperature obtained from the satellite image with the center wavelength of 10.4 μm at the estimation target point illustrated in FIG. 4 and the satellite image with the center wavelength of 12.3 μm illustrated in FIG. Switch the correction factor. FIG. 4 is an example of a satellite image with a center wavelength of 10.4 μm. FIG. 5 is an example of a satellite image with a center wavelength of 12.3 μm. The vertical axis in FIGS. 4 and 5 represents latitude and the horizontal axis represents longitude. In both FIGS. 4 and 5, the amount of reflected light is represented by the shading indicated by the band located on the right side of the paper. Here, the correction coefficient is assumed to be a common value in a certain area. The fixed range area is, for example, a range such as Kanto, Chubu, and Kansai.

After that, the solar radiation estimated value calculation unit 101 causes the data storage unit 102 to store the calculated solar radiation amount estimated value.

The data storage unit 102 is a storage device such as a hard disk. The data storage unit 102 stores weather data such as temperature, air pressure, wind speed, humidity, and amount of solar radiation provided by the weather information distribution system 3, and accumulates the data in association with the measurement date and time. In addition, the data storage unit 102 stores the estimated solar radiation amount calculated by the estimated solar radiation calculation unit 101, and accumulates each estimated solar radiation amount corresponding to the target day. Further, the data storage unit 102 acquires the predicted solar radiation amount calculated by the numerical weather prediction execution unit 131 and the predicted solar radiation calculated by the cloud movement prediction execution unit 132, which will be described later, and associates them with the prediction target day. accumulate.

The predicted value calculation unit 103 calculates a predicted solar radiation amount value for a prediction target date and time at a prediction target point using a plurality of prediction methods. The predicted value calculation unit 103 has a numerical weather prediction execution unit 131 and a cloud movement prediction execution unit 132 .

The numerical weather prediction execution unit 131 acquires weather data from the data storage unit 102 . Then, the numerical weather prediction executing unit 131 repeats the simulation by numerically expressing the weather phenomenon using the acquired weather data, thereby calculating the predicted amount of solar radiation. After that, the numerical weather prediction execution unit 131 stores the calculated solar radiation amount prediction value in the data storage unit 102 for accumulation. Below, the solar radiation amount prediction value calculated|required by the numerical weather forecast execution part 131 is called "solar radiation amount numerical prediction value." This solar radiation amount numerical prediction value corresponds to an example of a "first prediction value", and the numerical weather prediction execution unit 131 corresponds to an example of a "first prediction value calculation unit".

The cloud movement prediction execution unit 132 acquires satellite image data from the data storage unit 102 . Then, the cloud movement prediction execution unit 132 estimates the moving direction and speed of the cloud from the time difference of the satellite images. Next, the cloud movement prediction execution unit 132 predicts the cloud distribution several hours later using the estimated cloud movement direction and speed. This prediction method is sometimes called cloud motion vector prediction. After that, the cloud movement prediction executing unit 132 calculates a predicted solar radiation amount using the predicted cloud distribution. The cloud movement prediction execution unit 132 stores and accumulates the calculated solar radiation amount prediction value in the data storage unit 102 . Below, the solar radiation amount prediction value calculated|required by the cloud movement prediction execution part 132 is called "solar radiation amount cloud movement prediction value." This solar radiation amount cloud movement predicted value corresponds to an example of a "second predicted value", and the cloud movement prediction execution unit 132 corresponds to an example of a "second predicted value calculation unit".

The prediction formula generation unit 104 acquires the prediction target time and the prediction execution time specified by the operator. The prediction target time is the time for which the amount of solar radiation is to be predicted. The prediction execution time is the time at which the prediction is executed. For example, it is conceivable that a prediction with 12:00 as the prediction target time is performed at the prediction execution time of 9:00.

In addition, the prediction formula generating unit 104 has a template of a prediction formula for predicting the amount of solar radiation at the prediction target time in which the coefficients and constant terms are undetermined. For example, the prediction formula generation unit 104 can use the following formula (2) as a formula for predicting the solar radiation at the point s at 12:00.

Here, the input variable p when generating the solar radiation amount prediction formula is information that can be obtained by the time the solar radiation amount prediction formula is generated. As input variables, for example, observed solar radiation at the prediction target point at the prediction execution time, estimated solar radiation from satellite images, predicted solar radiation at the prediction target time by other methods such as numerical weather forecasting, A weather prediction value at the prediction target time at the prediction target point, such as temperature, can be used. In addition, in this embodiment, an estimated solar radiation amount at a prediction target time at a point other than the prediction target point, a solar radiation predicted value by numerical weather forecast, and the like are used as input variables.

P is the number of input variables, and the prediction formula generation unit 104 according to this embodiment can use multiple input variables in formula (2). Also, the prediction formula generation unit 104 determines coefficients and constant terms so as to match the past data. The method of determining the coefficients and the constant term is described below.

The prediction formula generation unit 104 determines coefficients and constant terms using, for example, the method of least squares. Specifically, the prediction formula generating unit 104 determines coefficients and constant terms so as to minimize the mean squared error defined for the past data represented by the following formula (3).

Here, i corresponds to a date, and for example, i=1 for one day before the prediction implementation date, i=2 for two days before, . . . , and i=N for N days before. Then, the prediction formula generating unit 104 uses the estimated solar radiation amount at the past prediction target time at the prediction target point calculated by the solar radiation estimated value calculation unit 101 as the solar radiation s and i at the prediction target time in Equation (3).

In the present embodiment, the prediction formula generation unit 104 also generates the estimated solar radiation amount at the past prediction execution time calculated by the solar radiation estimated value calculation unit 101 and the past prediction target time calculated by the numerical weather prediction execution unit 131. The numerical predicted value of insolation and the predicted value of insolation cloud movement at the prediction target time calculated by the cloud movement prediction execution unit 132 are used as input variables. Furthermore, the prediction formula generation unit 104 uses as input variables estimated solar radiation amounts, numerical predicted solar radiation amounts, and predicted solar cloud movement values at past prediction execution times at a plurality of points other than the prediction target point.

Then, when generating the solar radiation amount prediction formula, the prediction formula generation unit 104 acquires the input variables in the past fixed period from the data storage unit 102 . For example, the prediction formula generation unit 104 acquires data from 365 to 730 pieces before the latest data stored in the data storage unit 102 as data to be used as the input variable p. This means that the data from the latest data to 365 to 730 pieces before corresponds to data for about 1 to 2 years.

Thus, when using a plurality of input variables, the prediction formula generator 104 obtains the coefficient p for each input variable. Furthermore, by using data at a plurality of prediction target points as input variables, the prediction formula generation unit 104 acquires the coefficient p and the constant term for each prediction target point.

Then, the prediction formula generation unit 104 substitutes the calculated coefficient p and the constant term into the formula (2) to generate a solar radiation amount prediction formula for each prediction target point. After that, the prediction formula generation unit 104 outputs the generated solar radiation amount prediction formula to the prediction execution unit 105 . Here, in the present embodiment, the prediction formula generation unit 104 generated a prediction formula in advance using past data and provided the solar radiation amount prediction formula to the prediction execution unit 105, but the generation timing of the prediction formula is limited to this. do not have. For example, when the prediction execution unit 105 executes prediction, the prediction formula generation unit 104 may generate a solar radiation amount prediction formula using data up to that point, and provide the prediction execution unit 105 with the solar radiation amount prediction formula.

Information obtained by numerical weather forecast includes various predicted values such as temperature prediction, in addition to numerical predicted values of solar radiation. In this embodiment, the prediction formula generation unit 104 uses the numerical prediction value of solar radiation as an input variable, but the present invention is not limited to this, and other information obtained by numerical weather forecast may be used.

In addition, there are predicted values for a plurality of time zones in the numerical forecast value of solar radiation in the numerical weather forecast and the predicted value of cloud movement in solar radiation based on cloud moving speed estimation. In the present embodiment, the prediction formula generation unit 104 uses the numerical predicted value of solar radiation at the prediction target time and the predicted movement of clouds of insolation, but is not limited to this. In addition to movement prediction values, solar radiation amount prediction values for other time periods may also be used. In addition, there are predicted values for a plurality of points in the solar radiation amount prediction value of the numerical weather forecast and the solar radiation amount prediction value based on the cloud moving speed estimation. In this embodiment, the prediction formula generating unit 104 uses the predicted solar radiation amount of a plurality of points, but is not limited to this, and is limited to the predicted solar radiation amount of one point near the prediction target point or points around it. may be used.

The prediction execution unit 105 acquires the prediction target time and the prediction execution time specified by the operator. The prediction execution unit 105 also acquires the solar radiation amount prediction formula from the prediction formula generation unit 104 . Furthermore, the prediction execution unit 105 acquires the estimated amount of solar radiation at the prediction execution time from the data storage unit 102 . The prediction execution unit 105 also acquires from the data storage unit 102 the numerical prediction value of the solar radiation amount and the predicted cloud movement value of the solar radiation amount at the prediction target time. In this embodiment, the estimated solar radiation amount at the prediction execution time, the numerical prediction value of solar radiation amount at the prediction target time, and the cloud movement prediction value of solar radiation amount are input variables. The prediction execution unit 105 acquires these input variables for each prediction target point.

Then, the prediction executing unit 105 uses the acquired input variables in Expression (2) to calculate the predicted amount of solar radiation at the prediction target time for each prediction target point. Then, the prediction execution unit 105 outputs the solar radiation amount prediction value for each prediction target point to the notification unit 106 .

Here, the prediction execution unit 105 may further use other information obtained from numerical weather forecasts as input variables in addition to the numerical prediction value of the amount of solar radiation according to the prediction variables used by the prediction formula generation unit 104 . In addition, the prediction execution unit 105 generates the numerical prediction value of solar radiation amount at the prediction target time and the predicted cloud movement value of solar radiation amount according to the solar radiation amount prediction value used by the prediction formula generation unit 104, and calculates the amount of solar radiation in other time zones. A predicted value may be used.

The notification unit 106 receives an input of the solar radiation amount prediction value of each prediction target point from the prediction execution unit 105 . And the notification part 106 transmits the solar radiation amount prediction value of each prediction object point to the terminal device 4, and notifies it to a user.

Here, with reference to FIG. 6, the outline|summary of the solar radiation amount prediction process by the solar radiation prediction apparatus 1 which concerns on a present Example is demonstrated collectively. FIG. 6 is a diagram showing an overview of the solar radiation amount prediction processing by the solar radiation prediction device according to the first embodiment.

A process P1 surrounded by a dashed-dotted line frame in FIG. 6 represents a solar radiation amount prediction formula generation process. Processing P2 surrounded by a dashed frame represents the solar radiation amount prediction processing at the prediction target time.

The prediction formula generating unit 104 of the solar radiation prediction device 1 acquires the past data 201 of a fixed period in the past for each prediction target point. The prediction formula generation unit 104 includes, as past data 201, an estimated solar radiation amount 211 at a past prediction execution time, an estimated solar radiation amount 212 at a past prediction target time, a numerical predicted solar radiation amount 213 at a past prediction target time, and a past The cloud movement prediction value 214 for the amount of solar radiation at the prediction target time is acquired. Then, the prediction formula generation unit 104 generates the solar radiation amount prediction formula 203 for each prediction target point using the acquired past data.

The prediction executing unit 105 of the solar radiation prediction device 1 acquires, as the input variable 202, the solar radiation amount prediction value 221 at the actual prediction execution time at each prediction target point. Furthermore, the prediction executing unit 105 acquires, as input variables 202, a solar radiation amount numerical prediction value 222 and a solar radiation amount cloud movement prediction value 223 at the actual prediction execution time at each prediction target point. Then, the prediction execution unit 105 inputs the acquired input variables 202 to the solar radiation amount prediction formula for each prediction target point generated by the prediction formula generation unit 104 to calculate the solar radiation amount prediction value for each prediction target point. Thereby, the prediction execution part 105 calculates the solar radiation amount prediction value in each of the points 204 of FIG.

Next, with reference to FIG. 7, the flow of processing for generating a solar radiation amount prediction formula by the solar radiation prediction device 1 according to the present embodiment will be described. FIG. 7 is a flowchart of processing for generating a solar radiation amount prediction formula by the solar radiation prediction device according to the first embodiment.

The solar radiation estimated value calculator 101 acquires a satellite image from the geostationary meteorological satellite 2 (step S101).

Next, the solar radiation estimated value calculation unit 101 calculates the solar radiation estimated value at the time when the satellite image data was captured by obtaining the attenuation of extra-atmospheric solar radiation from the acquired satellite image (step S102). After that, the solar radiation estimated value calculation unit 101 transmits the calculated solar radiation amount estimated value to the data storage unit 102 .

Also, the predicted value calculation unit 103 acquires the weather data provided from the weather information distribution system 3 from the data storage unit 102 (step S103).

The numerical weather forecast executing unit 131 calculates the numerical forecast value of the amount of solar radiation at the specified prediction target time by numerical weather forecasting using the weather data (step S104). After that, the numerical weather forecast execution unit 131 transmits the calculated numerical forecast value of solar radiation to the data storage unit 102 .

On the other hand, the cloud movement prediction executing unit 132 calculates the predicted amount of solar radiation and cloud movement at the specified prediction target time using the weather data and the cloud movement model (step S105). After that, the cloud movement prediction execution unit 132 transmits the calculated cloud movement prediction value to the data storage unit 102 .

The data storage unit 102 stores the estimated solar radiation amount calculated by the estimated solar radiation calculation unit 101, the numerical predicted solar radiation amount calculated by the numerical weather prediction execution unit 131, and the solar radiation calculated by the cloud movement prediction execution unit 132. Receive predicted cloud movement. Then, the data storage unit 102 accumulates the acquired solar radiation estimated value, solar radiation numerical prediction value, and solar radiation cloud movement prediction value (step S106). Here, in the flowchart of FIG. 7, for convenience of explanation, the solar radiation prediction device 1 is explained to acquire the estimated solar radiation amount, the numerical predicted solar radiation amount, and the predicted cloud movement amount of solar radiation through a series of processes. These acquisition timings may be different.

The prediction formula generation unit 104 determines whether or not to execute prediction formula generation, for example, based on whether or not a predetermined periodic prediction formula generation timing has arrived (step S107). If prediction formula generation is not to be executed (step S107: NO), the process returns to step S101.

On the other hand, if the prediction formula is to be generated (step S107: affirmative), the prediction formula generation unit 104 acquires from the data storage unit 102 the estimated amount of solar radiation at the prediction execution time for a certain period of time in the past (step S108).

Further, the prediction formula generation unit 104 acquires the estimated solar radiation amount, the numerical predicted solar radiation amount, and the predicted cloud movement of solar radiation amount at the prediction target time in the past certain period (step S109).

Then, the prediction formula generation unit 104 determines coefficients and constant terms in formula (3) by the method of least squares using the acquired data (step S110).

After that, the prediction formula generation unit 104 substitutes the determined coefficients and constant terms into the formula (3) to generate a solar radiation amount prediction formula for each prediction target point (step S111).

Next, with reference to FIG. 8, the flow of solar radiation amount prediction processing by the solar radiation prediction device 1 according to the present embodiment will be described. FIG. 8 is a flowchart of solar radiation amount prediction processing by the solar radiation prediction device according to the first embodiment.

The solar radiation estimated value calculation unit 101 acquires the satellite image at the prediction implementation time from the geostationary meteorological satellite 2 (step S201).

Next, the solar radiation estimated value calculation unit 101 calculates the solar radiation estimated value at the prediction implementation time by obtaining the attenuation of the extra-atmospheric solar radiation from the acquired satellite image (step S202). After that, the solar radiation estimated value calculation unit 101 transmits the calculated solar radiation amount estimated value to the data storage unit 102 .

Also, the predicted value calculation unit 103 acquires the weather data provided from the weather information distribution system 3 from the data storage unit 102 (step S203).

The numerical weather forecast executing unit 131 calculates the numerical forecast value of the amount of solar radiation at the actual prediction target time by numerical weather forecasting using the weather data acquired by the forecast execution time (step S204). After that, the numerical weather forecast execution unit 131 stores the calculated numerical forecast value of solar radiation in the data storage unit 102 .

On the other hand, the cloud movement prediction executing unit 132 calculates a predicted cloud movement amount of solar radiation at the actual prediction target time using weather data and a cloud movement model (step S205). After that, the cloud movement prediction execution unit 132 stores the calculated cloud movement prediction value in the data storage unit 102 .

The prediction execution unit 105 acquires the solar radiation amount prediction formula for each prediction target point from the prediction formula generation unit 104 . The prediction execution unit 105 also acquires the estimated solar radiation amount at the actual prediction execution time from the data storage unit 102 . Furthermore, the prediction executing unit 105 acquires the numerical prediction value of the solar radiation amount and the predicted cloud movement value of the solar radiation amount at the actual prediction target time from the data storage unit 102 . The prediction execution unit 105 uses the acquired solar radiation estimated value, solar radiation numerical prediction value, and solar radiation cloud movement prediction value as input variables. Then, the prediction execution unit 105 substitutes each input variable into the solar radiation amount prediction formula at each prediction target point (step S206).

Then, the prediction execution unit 105 calculates the predicted solar radiation amount at the prediction target time at each prediction target point using the solar radiation prediction formula in which the input variable is substituted, and predicts the solar radiation amount at the prediction target time at each prediction target point. is executed (step S207). After that, the prediction execution unit 105 outputs the solar radiation amount prediction value for each prediction target point to the notification unit 106 .

The notification unit 106 receives an input of the solar radiation amount prediction value for each prediction target point from the prediction execution unit 105 . And the notification part 106 notifies a user of the solar radiation amount prediction value in each prediction object point via the terminal device 4 (step S208).

As described above, the solar radiation prediction apparatus according to the present embodiment generates a solar radiation amount prediction formula using past data representing predetermined input variables. For example, the solar radiation forecasting device can estimate the solar radiation amount at the prediction execution time and the forecast target time calculated from the meteorological satellite image, the numerical forecast value of the solar radiation amount by numerical weather prediction, and the forecast value of the solar radiation cloud movement by cloud movement vector prediction. A solar radiation prediction formula is generated using past data at multiple locations as input variables. Then, the solar radiation prediction apparatus according to the present embodiment uses the generated solar radiation amount prediction formula with the input variables at the time of prediction execution to determine the solar radiation amount prediction value at the prediction target time. For example, the solar radiation prediction device uses the estimated solar radiation amount at the prediction execution time, the numerical predicted solar radiation amount at the prediction target time, and the predicted solar radiation amount cloud movement value by cloud movement vector prediction at multiple points . Determine volume predictions.

By predicting the amount of solar radiation using information on the amount of solar radiation at multiple points in this way, it is possible to take into account the possibility that clouds may flow into the prediction target point from other points, thereby further improving the accuracy of the amount of solar radiation prediction. can. In addition, since the prediction results obtained by a plurality of prediction methods can be taken into consideration, the advantages of each prediction method can be taken into account, and the prediction errors in the long term and the short term can be reduced. In particular, it is possible to improve the prediction accuracy on days with large variations in solar radiation. Here, the solar radiation prediction by the solar radiation prediction device according to the present embodiment may be either the intensity of radiation (unit: W/m ² or the like) or the amount of radiation (unit: J/m ² or the like), and any information A similar effect can be obtained by using

Next, a solar radiation prediction device according to a second embodiment will be described. The solar radiation prediction apparatus according to this embodiment differs from that of the first embodiment in that the calculation time is reduced by lowering the resolution. The solar radiation prediction device according to this embodiment is also represented by the block diagram of FIG. In the following description, descriptions of the functions of the same units as in the first embodiment will be omitted.

Solar radiation estimates using Himawari-8 satellite image data and the like are data with high spatial resolution, and solar radiation estimates for a huge number of points can be obtained. For example, a 1 km resolution solar radiation estimate for the Kanto region (34.50 to 37.25 degrees latitude and 138.13 to 140.99 degrees longitude) includes data for 73,600 points. When generating the solar radiation amount prediction formula for each of these points, depending on the computer performance, it may take a huge amount of calculation time to generate the solar radiation amount prediction formula.

Therefore, the solar radiation prediction device 1 according to the present embodiment thins out the prediction points to lower the resolution of the prediction target points and reduce the number of prediction formulas to be generated. The processing for reducing the resolution of the predicted point will be described below.

The solar radiation estimated value calculation unit 101 receives an input specifying a point to be selected in advance from the operator. For example, the solar radiation estimated value calculation unit 101 receives an input of an instruction to select points at intervals of 16 km from each point with a resolution of 1 km. In this case, the predicted points are represented as respective points 400 shown in FIG. 9, for example, in the Kanto region. That is, the solar radiation estimated value calculation unit 101 calculates the solar radiation amount estimated value of each point 400 in FIG. FIG. 9 is a diagram showing prediction target points when the resolution is reduced.

In this case, the numerical weather prediction execution unit 131 and the cloud movement prediction execution unit 132 also calculate the solar radiation numerical prediction value and the solar radiation amount cloud movement prediction value at the same point as the solar radiation estimated value calculation unit 101, thereby lowering the resolution. may be performed.

The prediction formula generation unit 104 uses the solar radiation estimated value, the solar radiation numerical prediction value, and the solar radiation cloud movement prediction value stored in the data storage unit 102 to generate a low-resolution solar radiation prediction formula for each prediction target point. Generate.

The prediction execution unit 105 uses the low-resolution solar radiation prediction formula for each prediction target point stored in the data storage unit 102, as well as the solar radiation estimated value, the solar radiation numerical prediction value, and the solar radiation cloud movement prediction value. , a low-resolution solar radiation amount prediction value for each prediction target point is calculated.

Here, the prediction execution unit 105 can also correct the solar radiation prediction value to the resolution before the resolution reduction using the neighborhood method. Specifically, the prediction execution unit 105 determines, among the points having the resolution before the resolution reduction, the points for which the solar radiation amount prediction value cannot be obtained when the resolution is reduced, to the points having the solar radiation amount prediction value. Prediction is performed using the same solar radiation amount prediction value as that of one nearby point.

As described above, the solar radiation prediction apparatus according to the present embodiment can reduce the calculation time required for solar radiation amount prediction by reducing the resolution.

Next, a solar radiation prediction device according to Example 3 will be described. The solar radiation prediction device according to this embodiment differs from that of the first embodiment in that overlearning is suppressed. The solar radiation prediction device according to this embodiment is also represented by the block diagram of FIG. In the following description, descriptions of the functions of the same units as in the first embodiment will be omitted.

When a large number of input variables are used, there is a risk of over-learning in which a prediction formula that is excessively adapted to past data is generated. Therefore, in order to suppress such over-learning, the prediction formula generation unit 104 of the solar radiation prediction device 1 can use a model selection method such as AIC (Akaike's Information Criterion) or a regularization method such as Lasso regression.

As shown in FIG. 10, the prediction formula generation unit 104 according to the present embodiment reduces overfitting by reducing a large number of input variables to a small number of intermediate variables and generating a prediction formula using the sum of the intermediate variables. Suppress. FIG. 10 is a conceptual diagram of a technique for suppressing overlearning using intermediate variables according to the third embodiment. By using intermediate variables, information useful for prediction is extracted to the intermediate variables, while unnecessary information that becomes noise for prediction is removed. As a result, it is possible to prevent the prediction formula from being excessively adapted to the past data, and it is possible to perform highly accurate prediction using the generated prediction formula.

The prediction formula generation unit 104 determines the coefficients and the constant term of the prediction formula using the following formulas (4) and (5) when determining the coefficient p and the constant term of the formula (3).

Q is the number of intermediate variables. The prediction formula generation unit 104 uses this intermediate variable q to determine coefficients and constant terms in the prediction formula by, for example, Partial Least Squares Regression.

Here, in prediction using intermediate variables, the number of intermediate variables is determined in advance. The number of intermediate variables is determined based on past prediction results. For example, assuming that the number of intermediate variables is 1, 2, . It is possible to find the number of intermediate variables with the smallest prediction error for the result. Then, the prediction formula generating unit 104 uses the obtained number of intermediate variables as the number of intermediate variables used for generating the prediction formula.

Also, the prediction execution unit 105 according to this embodiment also executes prediction using intermediate variables. Here, the prediction execution unit 105 may use the same number of intermediate variables as the number of intermediate variables used by the prediction formula generation unit 104 when generating the prediction formula. In addition, the prediction execution unit 105 may calculate the prediction error for each case where the intermediate variables are changed separately from the prediction formula generation unit 104 and set the number of intermediate variables to a value different from that of the prediction formula generation unit 104 .

As described above, the solar radiation prediction apparatus according to the present embodiment uses intermediate variables to calculate the coefficients and constant terms of the prediction formula. As a result, over-learning can be suppressed, and more accurate solar radiation amount prediction can be performed.

Here, the accuracy of solar radiation prediction by the solar radiation prediction device 1 according to the present embodiment will be described with reference to FIG. 11 . FIG. 11 is a diagram for explaining the accuracy of solar radiation prediction by the solar radiation prediction device. The vertical axis of FIG. 11 represents the solar radiation intensity, and the horizontal axis represents the date and time of the prediction target. FIG. 11 shows the prediction result of the amount of solar radiation at a specific point when 9:00 is the prediction implementation time and 10:00, 11:00, .

In this case, at 6 o'clock every day, the solar radiation prediction device 1 determines the coefficients and constant terms of the prediction formula for each prediction target point and each prediction target time from past data for the nearest one year. Furthermore, here, the solar radiation prediction device 1 uses a prediction target point obtained by lowering the resolution of 1 km obtained from satellite image data to 16 km. In addition, the solar radiation prediction device 1 suppresses overlearning using intermediate variables. The number of intermediate variables in the partial least squares regression is the same value for all the prediction target points, and is different depending on the prediction implementation time and the prediction target time. In this case, the prediction formula is represented by the following formula (6).

In Expression (6), τ is the prediction implementation time, and t is the prediction target time. In addition, the estimated solar radiation amount rτ is the estimated solar radiation amount at the point r at the prediction execution time at the time τ. In addition, the solar radiation amount numerical prediction value jt is the solar radiation amount prediction value based on the numerical weather prediction at the point j. In addition, the solar radiation amount cloud movement predicted value kt is the solar radiation amount predicted value based on the cloud movement vector prediction at the point k. Here, as a cloud movement vector prediction method, the solar radiation prediction method of NuWFAS developed by the Central Research Institute of Electric Power Industry was used. From NuWFAS, predicted values of insolation cloud movement at points spaced by 4 km are obtained.

Then, the solar radiation prediction device 1 inputs the solar radiation estimated value at the prediction execution time estimated from the satellite image obtained at 9:00 just before 9:00, the solar radiation numerical value predicted at the prediction target time, and the solar radiation cloud movement predicted value. A solar radiation prediction value is calculated by substituting it as a variable into the solar radiation prediction formula.

The prediction result by the solar radiation prediction device 1 in this case is represented by the graph 302 in FIG. Also, the actual observed value of the amount of solar radiation at the actual specific point in this case is represented by a graph 301 . In FIG. 11 , the time period during which the solar radiation amount prediction value is not calculated is omitted. Thus, it can be said that the prediction result of the solar radiation amount using the solar radiation prediction apparatus 1 according to the present embodiment approximates the actual observation result to some extent, and it can be said that the solar radiation amount can be predicted with high accuracy.

1 solar radiation prediction device 2 geostationary meteorological satellite 3 weather information distribution system 4 terminal device 10 solar radiation prediction system 101 solar radiation estimated value calculation unit 102 data storage unit 103 predicted value calculation unit 104 prediction formula generation unit 105 prediction execution unit 106 notification unit 131 numerical weather Prediction execution unit 132 Cloud movement prediction execution unit

Claims (10)

A first solar radiation estimated value calculation unit that calculates an estimated solar radiation value at a prediction target time from a satellite image;
Estimated solar radiation at past prediction execution time calculated based on the satellite image, first solar radiation forecast calculated by numerical weather forecast based on weather information, and forecasting cloud movement based on the weather information Prediction formula generation for generating a prediction formula based on the estimated solar radiation at the prediction target time in the past calculated by the first estimated solar radiation calculation unit, using the second predicted solar radiation calculated as an input variable. Department and
A solar radiation prediction device, comprising: a prediction execution unit that performs solar radiation prediction at the prediction target time using the prediction formula generated by the prediction formula generation unit. 2. The solar radiation prediction apparatus according to claim 1, further comprising a second estimated solar radiation value calculation unit that calculates an estimated solar radiation value at said prediction execution time based on said satellite image. 3. The solar radiation prediction device according to claim 1, further comprising a first predicted value calculation unit that obtains the first solar radiation predicted value by the numerical weather prediction based on the weather information. The solar radiation according to any one of claims 1 to 3, further comprising a second predicted value calculation unit for predicting the cloud movement based on the weather information and obtaining the second solar radiation predicted value. prediction device. The prediction formula generation unit uses the estimated solar radiation value, the first predicted solar radiation value , and the second predicted solar radiation value at the prediction execution time at another point to generate the prediction formula for a specific point. The solar radiation prediction device according to any one of claims 1 to 4, characterized in that: The prediction formula generation unit calculates the solar radiation estimated values at the past prediction execution time at the plurality of points according to the spatial resolution of the satellite image, and the plurality of points calculated based on the actual measurement values of the weather information at the plurality of points. and the second predicted solar radiation value at the plurality of locations calculated based on the measured values of the weather information at the plurality of locations as the input variables , and for each of the plurality of locations the generate a prediction formula,
The solar radiation prediction device according to any one of claims 1 to 5, wherein the prediction execution unit performs the solar radiation prediction using the prediction formula for each of the plurality of points. The prediction formula generation unit selects some points from among the multiple points from which the estimated solar radiation value is obtained as at least the upper limit of the spatial resolution of the satellite image, and the selected points are the plurality of points. 7. The solar radiation prediction device according to claim 6, wherein the prediction formula for the plurality of points is generated using estimated solar radiation values as the input variables. The prediction formula generation unit aggregates the estimated solar radiation value at the prediction execution time in the past, the first predicted solar radiation value, and the second predicted solar radiation value into intermediate variables, and calculates the prediction formula based on a linear combination of the intermediate variables. The solar radiation prediction device according to any one of claims 1 to 7, characterized in that it generates . Calculate the estimated solar radiation at the forecast target time based on the satellite image,
Estimated solar radiation at the past prediction execution time calculated from the satellite image, the first solar radiation forecast calculated by numerical weather forecast based on weather information, and cloud movement predicted and calculated based on the weather information generating a prediction formula based on the estimated solar radiation value at the prediction target time in the past using the second predicted solar radiation value as an input variable ;
A solar radiation prediction method, comprising: performing solar radiation prediction at the prediction target time using the generated prediction formula. Calculate the estimated solar radiation at the forecast target time based on the satellite image,
Estimated solar radiation at the past prediction execution time calculated from the satellite image, the first solar radiation forecast calculated by numerical weather forecast based on weather information, and cloud movement predicted and calculated based on the weather information generating a prediction formula based on the estimated solar radiation value at the prediction target time in the past using the second predicted solar radiation value as an input variable ;
A solar radiation prediction program for causing a computer to execute a process of performing solar radiation prediction at the prediction target time using the generated prediction formula.

Images