Solar power generation prediction device and solar power generation prediction method
The solar power generation forecasting device uses ensemble forecasting and multiple mesh data types to accurately predict snow accumulation and adjust power generation forecasts, addressing inaccuracies in existing methods by incorporating machine learning for enhanced precision.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- HIATACHI POWER SOLUTIONS CO LTD
- Filing Date
- 2024-11-15
- Publication Date
- 2026-05-27
Smart Images
Figure 2026087351000001_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to a solar power generation amount prediction device and a solar power generation amount prediction method.
Background Art
[0002] Regarding the prediction of the power generation amount considering snow accumulation on a solar power generation panel, for example, the technique described in Patent Document 1 is known. That is, Patent Document 1 describes "calculating an actual snow accumulation coefficient based on the first power generation output of the solar power generation facility when there is no snow accumulation and the second power generation output of the solar power generation facility when there is snow accumulation".
Prior Art Documents
Patent Documents
[0003]
Patent Document 1
Summary of the Invention
Problems to be Solved by the Invention
[0004] In the technique described in Patent Document 1, although it is considered that the power generation amount of the solar power generation panel decreases due to the influence of snow accumulation, the possibility that the weather forecast regarding snow accumulation is incorrect is not particularly considered, and there is room for further improvement in accuracy.
[0005] Therefore, an object of the present disclosure is to provide a solar power generation amount prediction device or the like that can accurately predict the power generation amount of a solar power generation facility.
Means for Solving the Problems
[0006] To solve the aforementioned problems, this disclosure provides a solar power generation forecasting device comprising: a weather parameter acquisition unit that acquires predicted values of predetermined weather parameters as multiple types of mesh data with different mesh sizes on a map, and which acquires the multiple types of mesh data as a set of multiple numerical forecasts with different initial values when forecasting the weather parameters, the device further comprising: a weather parameter variation calculation unit that calculates the variation of the predicted values of the weather parameters in the numerical forecast for each of the multiple types of mesh data; a forecast mode selection unit that selects one of a plurality of forecast modes based on the variation of the predicted values of the weather parameters; a snow depth forecasting unit that predicts the amount of snow in the area where the solar power generation facility is located based on the predetermined forecast mode selected by the forecast mode selection unit; a power generation reduction rate calculation unit that calculates the power generation reduction rate of the solar power generation facility based on the predicted value of the snow depth; and a power generation forecasting unit that predicts the amount of power generated by the solar power generation facility based on the power generation reduction rate. [Effects of the Invention]
[0007] According to this disclosure, it is possible to provide a solar power generation prediction device, etc., that can predict the amount of power generated by a solar power generation facility with high accuracy. [Brief explanation of the drawing]
[0008] [Figure 1] This is a diagram showing the configuration of a solar power generation prediction device according to the first embodiment. [Figure 2A] This is an explanatory diagram showing an example of a solar power generation prediction device according to the first embodiment, where the variation of the ensemble members is small. [Figure 2B] This is an explanatory diagram showing an example of a case where there is a large variation in the ensemble members in the solar power generation prediction device according to the first embodiment. [Figure 3] This is a functional block diagram of the solar power generation prediction device according to the first embodiment. [Figure 4] This flowchart shows the processing performed by the processing unit of the solar power generation prediction device according to the first embodiment. [Figure 5] It is a flowchart regarding the selection of the prediction mode of the solar power generation amount prediction device according to the first embodiment. [Figure 6] It is a flowchart regarding the first prediction mode of the solar power generation amount prediction device according to the first embodiment. [Figure 7] It is a flowchart regarding the second prediction mode of the solar power generation amount prediction device according to the first embodiment. [Figure 8] It is a flowchart regarding the third prediction mode of the solar power generation amount prediction device according to the first embodiment. [Figure 9] It is a configuration diagram including the solar power generation amount prediction device according to the second embodiment. [Figure 10] It is a functional block diagram of the solar power generation amount prediction device according to the second embodiment. [Figure 11] It is a flowchart showing the processing executed by the processing unit of the solar power generation amount prediction device according to the second embodiment. [Figure 12] It is a functional block diagram of the solar power generation amount prediction device according to the third embodiment. [Figure 13] It is a flowchart showing the processing executed by the processing unit of the solar power generation amount prediction device according to the third embodiment. [Figure 14] It is an explanatory diagram showing an example of a pseudo-ensemble member acquired by the solar power generation amount prediction device according to the modification example.
Mode for Carrying Out the Invention
[0009] ≪First Embodiment≫ <Configuration of Solar Power Generation Amount Prediction Device> FIG. 1 is a configuration diagram including the solar power generation amount prediction device 10 according to the first embodiment. The solar power generation amount prediction device 10 is a device that predicts the power generation amount of the solar power generation facility 40. The above-mentioned "power generation amount" may be the generated electric energy amount ([kWh]) in each time zone, or may be the generated electric power ([kW]) at each moment. The solar power generation facility 40 is a facility that converts the light energy of sunlight into electric energy, and includes a solar power generation panel (not shown).
[0010] As shown in FIG. 1, the photovoltaic power generation amount prediction device 10 includes a storage unit 11, a processing unit 12, an input / output interface 13, and a communication interface 14, which are configured to be connected to each other via an internal bus 15 in a predetermined manner. Note that FIG. 1 shows an example in which the photovoltaic power generation amount prediction device 10 is configured as one device, but it is not limited thereto. For example, the functions of the photovoltaic power generation amount prediction device 10 may be distributed among a plurality of computers such as a cloud server and an edge server.
[0011] In the storage unit 11 of the photovoltaic power generation amount prediction device 10, in addition to a predetermined program and data being stored in advance, the processing results of the processing unit 12 are appropriately stored. Such a storage unit 11 includes, although not shown, a volatile memory such as a RAM (Random Access Memory) and a register, and a non-volatile memory such as a ROM (Read Only Memory) and an HDD (Hard Disk Drive).
[0012] The processing unit 12 executes a predetermined process based on the data in the storage unit 11 and the data input via the input / output interface 13 and the communication interface 14. As such a processing unit 12, a processor such as a CPU (Central Processing Unit) is used. Then, the program stored in the ROM or HDD is read out and expanded in the RAM, and the processor executes various processes.
[0013] The input / output interface 13 is an interface used for data input from the input device 20 and data output to the display device 30. The input device 20 is, for example, a keyboard or a mouse, and is used when the user inputs data. The display device 30 is, for example, a display. Note that the input and output of data may be performed using a mobile terminal such as a smartphone or a tablet.
[0014] The communication interface 14 is an interface used for communication via the network N1. In the example shown in Figure 1, the solar power generation forecasting device 10 is connected to the solar power generation equipment 40 and also to the weather information server 50 via the network N1. The weather information server 50 is a server that provides predetermined weather information. The weather information may be data from the Japan Meteorological Agency, or it may be data from a predetermined information source other than the Japan Meteorological Agency.
[0015] <About weather forecasts> First, let me briefly explain the weather forecasts made by the Japan Meteorological Agency (JMA) and other organizations (i.e., the data stored in the weather information server 50). The JMA and other organizations make weather forecasts based on predetermined observation data and numerical weather prediction data. The aforementioned "observation data" consists of measured values of predetermined meteorological parameters (temperature, humidity, precipitation, wind direction, wind speed, etc.) obtained from ground-based weather observation stations and weather radar, as well as from weather satellites and oceanographic research vessels. The "numerical weather prediction data" consists of predicted values of meteorological parameters in each part (called a mesh) when the target area for weather forecasting is divided three-dimensionally at predetermined intervals in the latitude, longitude, and height directions, and is calculated by a computer (not shown) of the JMA or other organizations based on the observation data, etc.
[0016] In numerical weather prediction by the Japan Meteorological Agency and other organizations, a value called an "initial value" is first calculated for each mesh. Here, the "initial value" is the value of the meteorological parameter at the initial time that serves as the basis for weather forecasting using a predetermined numerical weather prediction model. Such initial values are calculated by integrating observational data with past prediction values (so-called data assimilation). Since the observational data obtained at a given observation point has low spatial resolution, errors are reduced by appropriately reflecting past prediction values based on numerical weather prediction.
[0017] Then, with initial values associated with each mesh, predicted values for meteorological parameters in each mesh are calculated by a computer (not shown) of the Japan Meteorological Agency or similar organization based on a predetermined numerical weather prediction model. This weather forecasting process is repeated at predetermined intervals, and the results are sequentially stored in the weather information server 50.
[0018] <About Ensemble Forecasts> For example, if one initial value is set for each mesh and a single prediction (so-called deterministic prediction) is made based on a predetermined numerical weather prediction model, there is a high probability that an error will occur between the predicted value and the actual value. This error is due to factors such as an insufficient amount of observational data, uncertainty in atmospheric behavior, and limitations of the numerical weather prediction model, and tends to increase over time. Therefore, the Japan Meteorological Agency and other organizations employ a method called "ensemble forecasting."
[0019] "Ensemble forecasting" is a method of making multiple numerical forecasts by slightly perturbing the initial state (i.e., initial values), and then statistically processing the results to make a probabilistic prediction that takes uncertainty into account. When multiple numerical forecasts are made based on several slightly different initial values, each individual prediction result is called an "ensemble member." In the first embodiment, the solar power generation forecasting device 10 predicts the amount of snowfall in the target area based on information from multiple ensemble members provided by the weather information server 50 (see Figure 1).
[0020] Figure 2A is an explanatory diagram illustrating an example where there is little variation among the ensemble members. In Figure 2A, the horizontal axis represents time, and the vertical axis represents the value of a predetermined meteorological parameter. The time t0 shown in Figure 2A is the initial time. The values of the meteorological parameters at this time t0 are the initial values described above. The "control run," shown as a thick line in Figure 2A, refers to the numerical forecast calculated without introducing any artificial errors into the initial values. The control run is also included in the ensemble members described above.
[0021] In the example in Figure 2A, the other five ensemble members are generated with slightly different initial values, based on the initial values of the control run (the values of the weather parameters at time t0). Time t1 shown in Figure 2A is a predetermined target time for prediction, and is set as the end point of a predetermined period (times t0 to t1) that starts from the initial time (time t0). In the example in Figure 2A, the variation in the ensemble members at time t1 is relatively small, and the error between the control run values (thick line) and the actual values (stars) is also small.
[0022] Figure 2B is an explanatory diagram illustrating an example where there is a large variation among the ensemble members. In the example in Figure 2B, the variability of the ensemble members at time t1 (the time of prediction) is greater than in the case of Figure 2A. Such variability is often caused by limitations of the weather prediction model or the resolution of the mesh model such as MSM. Note that the greater the variability of the ensemble members at time t1, the greater the uncertainty of the prediction, and the smaller errors included in the analyzed values (control run) tend to increase over time. In the example in Figure 2B, the error between the control run (thick line) value and the actual value (star) at time t1 is considerably large.
[0023] <About Mesh Data> As mentioned above, numerical weather forecasts by the Japan Meteorological Agency and other organizations associate predetermined meteorological parameter values with each mesh divided in the directions of latitude, longitude, and altitude. Multiple mesh models are generally used depending on the mesh scale (the spacing between grid points). For example, GSM (Global Model: 1st Mesh Data) and MSM (Mesoscale Model: 2nd Mesh Data) are used as mesh models.
[0024] The Global Scheme (GSM) is a mesh model that covers the entire globe, with a horizontal grid spacing of approximately 13 km and 128 vertical layers. Such a GSM is particularly suitable for global forecasts. On the other hand, the Multi-Scheme (MSM) is a mesh model that covers a specific region defined by latitude and longitude, with a horizontal grid spacing of approximately 5 km and 96 vertical layers. In other words, the mesh used to define regions in the latitude, longitude, and altitude directions is smaller in the MSM than in the GSM. Furthermore, considering the scale of the region covered by the mesh model, the MSM, which covers a specific region, is smaller than the GSM, which covers the entire globe. Such an MSM is more suitable for finer-grained weather forecasts than the GSM.
[0025] In the first embodiment, as an example, we will describe a case in which the solar power generation forecasting device 10 acquires mesh data of an ensemble forecast based on GSM and another mesh data of an ensemble forecast based on MSM from the weather information server 50 (see Figure 1) via the network N1. The aforementioned "mesh data" is data to which predetermined meteorological parameter values are associated with each mesh. In particular, the "ensemble forecast mesh data" includes multiple initial values that are slightly different at the initial time, and multiple prediction results (predicted values of meteorological parameters for each time period of a predetermined period starting from the initial time) when numerical predictions are made from each initial value. In other words, the "ensemble forecast mesh data" includes multiple ensemble members as prediction results.
[0026] Figure 3 is a functional block diagram of the solar power generation prediction device 10. As shown in Figure 3, the memory unit 11 stores weather parameters 111 and threshold values 112. The weather parameters 111 are acquired as needed from the weather information server 50 (see Figure 1) via the network N1 (see Figure 1). The threshold value 112 is a threshold value related to the degree of variation among multiple ensemble members and is pre-set. This threshold value 112 is used in the selection of the forecast mode, which will be described later.
[0027] As shown in Figure 3, the processing unit 12 of the solar power generation prediction device 10 includes a weather parameter acquisition unit 121, a weather parameter variation calculation unit 122, a prediction mode selection unit 123, a snow depth prediction unit 124, a power generation reduction rate calculation unit 125, and a power generation prediction unit 126.
[0028] The weather parameter acquisition unit 121 acquires predicted values of predetermined weather parameters related to snow cover as multiple types of mesh data with different mesh sizes on the map (for example, GSM and MSM mesh data). Such weather parameters include the horizontal and vertical distribution of temperature in a predetermined area including the location of the solar power generation facility 40 (see Figure 1), as well as relative humidity, cumulative precipitation, and cloud cover. The aforementioned "cumulative precipitation" is a predicted value of the amount of rain that falls over a predetermined period (for example, every hour). "Cloud cover" is a predicted value of the percentage of the sky (the area above the prediction target) that is covered by clouds.
[0029] The aforementioned multiple types of mesh data are acquired as sets of multiple numerical weather forecasts (i.e., ensemble members) with different initial values for predicting meteorological parameters. Specifically, the meteorological parameter acquisition unit 121 acquires mesh data for GSM ensemble forecasts and mesh data for MSM ensemble forecasts, which have smaller mesh sizes than GSM.
[0030] The weather parameter variability calculation unit 122 calculates the variability of predicted values of weather parameters in numerical weather forecasts for each of the multiple types of mesh data. Specifically, the weather parameter variability calculation unit 122 calculates a value indicating the degree of variability of predicted values of weather parameters based on the mesh data of the GSM ensemble forecast. Similarly, the weather parameter variability calculation unit 122 calculates a value indicating the degree of variability of predicted values of weather parameters based on the mesh data of the MSM ensemble forecast.
[0031] The value indicating the degree of variation is calculated using a well-known method, for example, based on the standard deviation of the meteorological parameter values at a predetermined forecast time (the values at time t1 in Figures 2A and 2B). Furthermore, one value indicating the degree of variation in the predicted meteorological parameter values is calculated for both GSM and MSM.
[0032] The prediction mode selection unit 123 selects one of several prediction modes based on the variability of meteorological parameters. Here, "prediction mode" refers to the method (algorithm) used to predict snowfall and other factors in the area where the solar power generation facility 40 (see Figure 1) is located, and is pre-set. Details of the prediction modes will be described later.
[0033] The snow depth prediction unit 124 predicts the amount of snow in the area where the solar power generation facility 40 (see Figure 1) is located, based on a predetermined prediction mode selected by the prediction mode selection unit 123. The power generation reduction rate calculation unit 125 calculates the power generation reduction rate of the solar power generation equipment 40 (see Figure 1) based on the predicted amount of snowfall. The "power generation reduction rate" is the rate of decrease in the amount of power generated by the solar power generation equipment 40, with the case where there is no snowfall being used as the baseline (100%).
[0034] The power generation prediction unit 126 predicts the amount of power generated by the solar power generation equipment 40 (see Figure 1) based on the power generation reduction rate described above. Note that the more snow there is on the solar power generation panels (not shown), the less sunlight reaches the panels, resulting in a decrease in power generation.
[0035] Figure 4 is a flowchart showing the processes performed by the processing unit of the solar power generation forecasting device (see also Figure 3 as appropriate). Note that when "START" is selected in Figure 4, it is assumed that the mesh data for the ensemble forecast is already stored in the weather information server 50 (see Figure 1) for both the large-mesh GSM and the small-mesh MSM. In step S101, the processing unit 12 uses the weather parameter acquisition unit 121 to acquire predicted values of weather parameters from the weather information server 50 as multiple types of mesh data with different mesh sizes (weather parameter acquisition step). Specifically, the weather parameter acquisition unit 121 acquires information from the weather information server 50 that includes GSM (first mesh data) and MSM (second mesh data) which has a smaller mesh than GSM, as multiple types of mesh data. Examples of weather parameters acquired in step S101 include the horizontal and vertical distribution of temperature, as well as relative humidity, cumulative precipitation, and cloud cover.
[0036] It should be assumed that the area covered by the mesh data acquired in step S101 includes the area of the solar power generation facility 40 (see Figure 1). In this case, one mesh (or multiple meshes including its surroundings) near the ground surface that includes the area of the solar power generation facility 40 may be selected by the user's input operation as the target for acquiring weather parameters. However, for the vertical distribution of temperature, which is one of the weather parameters, multiple layers of meshes in the vertical direction that include the area of the solar power generation facility 40 shall be selected.
[0037] Furthermore, the meteorological parameters acquired as mesh data in step S101 are assumed to include multiple ensemble members based on the ensemble forecast using GSM (the same applies to MSM). The initial time of the ensemble forecast is specified, for example, by user operation via the input device 20 (see Figure 1). The length of the target period in the ensemble forecast is set appropriately according to the type of mesh model, such as 132 hours for GSM and 39 hours for MSM.
[0038] Next, in step S102, the processing unit 12 calculates the variation in the predicted values of the weather parameters using the weather parameter variation calculation unit 122 (weather parameter variation calculation step). Specifically, the processing unit 12 calculates a value indicating the variation in temperature among multiple ensemble members at the target time (time t1 in Figures 2A and 2B) when using GSM. Similarly, the processing unit 12 also calculates a value indicating the variation in temperature for each weather parameter when using MSM. The type of weather parameter (one or more types) to be used for variation calculation is set in advance.
[0039] In step S103, the processing unit 12 selects a forecast mode using the forecast mode selection unit 123 (forecast mode selection step). That is, the processing unit 12 selects a forecast mode based on the magnitude of the variation in weather parameters in GSM and MSM, respectively. Details of the process in step S103 will be described later.
[0040] In step S104, the processing unit 12 predicts the amount of snow using the snow depth prediction unit 124 (snow depth prediction step). That is, the processing unit 12 predicts the amount of snow in the area including the solar power generation facility 40 (snow depth in a predetermined mesh) based on the prediction mode selected in step S103. Details of the processing in step S104 will be described later.
[0041] In step S105, the processing unit 12 calculates the power generation reduction rate of the solar power generation equipment 40 using the power generation reduction rate calculation unit 125 (power generation reduction rate calculation step). That is, the processing unit 12 calculates the power generation reduction rate of the solar power generation equipment 40 for each time period based on a predetermined formula that includes the amount of snowfall calculated in step S104.
[0042] In step S106, the processing unit 12 corrects the power generation reduction rate based on the tilt angle of the solar power generation panel (not shown). The tilt angle of the solar power generation panel (tilt angle relative to the horizontal direction) is stored in the memory unit 11 in advance, associated with the identification information of the solar power generation equipment 40. The larger the tilt angle of the solar power generation panel, the easier it is for the accumulated snow to slide off the solar power generation panel under its own weight, and therefore the power generation reduction rate tends to be smaller.
[0043] In step S107, the processing unit 12 predicts the amount of power generated by the solar power generation equipment 40 using the power generation prediction unit 126 (power generation prediction step). Specifically, the processing unit 12 calculates a predicted value of power generation by multiplying the amount of power generated by the solar power generation equipment 40 assuming there is no snow by the value of (1 - power generation reduction rate). The predicted value of power generation calculated in this way is associated with the identification information of the solar power generation equipment 40 and stored in the storage unit 11. After performing the processing in step S107, the processing unit 12 terminates the series of processes (END).
[0044] Figure 5 is a flowchart for selecting the prediction mode (see also Figure 3 as needed). The series of processes shown in Figure 5 corresponds to step S103 (selection of prediction mode) in Figure 4. In step S1031, the processing unit 12 determines whether the variation in weather parameters in GSM is below the first threshold. Here, the "first threshold" is a threshold that serves as the criterion for the processing unit 12 to decide whether or not to select the first forecast mode, and is set in advance.
[0045] If, in step S1031, the variation in the GSM weather parameters is below the first threshold (S1031: Yes), the processing unit 12 proceeds to step S1032. In this case, even when using GSM, the variation in the predicted values of the weather parameters (i.e., the variation of the ensemble members) is small, so it is highly likely that the weather above the solar power generation facility 40 (see Figure 1) will be dominated by global phenomena.
[0046] Furthermore, if multiple types of meteorological parameters are obtained, such as horizontal temperature distribution, relative humidity, and cumulative precipitation, the determination process in step S1031 may be performed on all of them, or on some of them. When multiple types of meteorological parameters are used in the determination process in step S1031, for example, the processing unit 12 may proceed to step S1032 if the variation of all of these multiple types of meteorological parameters is below the first threshold. Alternatively, the processing unit 12 may proceed to step S1032 if the variation of at least one of the multiple types of meteorological parameters is below the first threshold. The same applies to the processing in step S1033, which will be described later.
[0047] In step S1032, the processing unit 12 selects a first prediction mode using the prediction mode selection unit 123. That is, if the variation in the predicted values of meteorological parameters in the GSM (first mesh data) is below a first threshold (S1031: Yes), the prediction mode selection unit 123 selects a first prediction mode (S1032). Here, the "first prediction mode" is a prediction mode that is executed when there is a high probability that the snow cover in the area to be predicted is dominated by a global phenomenon, such as a pressure pattern of high pressure in the west and low pressure in the east caused by the Siberian high pressure system. Details of the first prediction mode will be described later.
[0048] Furthermore, if the variation in the GSM weather parameters is greater than the first threshold in step S1031 (S1031: No), the processing unit 12 proceeds to step S1033. In step S1033, the processing unit 12 determines whether the variation in the weather parameters of the MSM is below the second threshold. Here, the "second threshold" is a threshold that serves as the criterion for the processing unit 12 to decide whether or not to select the second forecast mode, and is set in advance.
[0049] Furthermore, the relationship between the first and second thresholds described above is not particularly limited. Depending on the pre-set conditions, the second threshold may be greater than the first threshold, the first and second thresholds may be equal, or the second threshold may be smaller than the first threshold.
[0050] If, in step S1033, the variation in the weather parameters of the MSM is less than or equal to the second threshold (S1033: Yes), the processing unit 12 proceeds to step S1034. In step S1034, the processing unit 12 selects the second prediction mode using the prediction mode selection unit 123. That is, if the variation in the predicted values of weather parameters in GSM (first mesh data) is greater than the first threshold (S1031: No), and furthermore, the variation in the predicted values of weather parameters in MSM (second mesh data) is less than or equal to the second threshold (S1033: Yes), the prediction mode selection unit 123 selects the second prediction mode (S1034).
[0051] Here, the "second forecast mode" is a forecast mode that is executed when the weather in the forecast area is likely to be dominated by phenomena that contain uncertainties that are difficult to capture with the spatial resolution of GSM. For example, the second forecast mode is executed when there is a high probability of being affected by medium-scale phenomena such as low-pressure systems off the southern coast, as well as by coastal topography and the location of cloud edges (the edge of cloud areas). Further details on the second forecast mode will be described later.
[0052] Furthermore, if the variation in the weather parameters of the MSM is greater than the second threshold in step S1033 (S1033: No), the processing unit 12 proceeds to step S1035. In step S1035, the processing unit 12 selects the third prediction mode using the prediction mode selection unit 123. That is, if the variation in the predicted values of weather parameters in GSM (first mesh data) is greater than the first threshold (S1031: No), and furthermore, the variation in the predicted values of weather parameters in MSM (second mesh data) is greater than the second threshold (S1033: No), the prediction mode selection unit 123 selects the third prediction mode (S1035).
[0053] Here, the "third forecast mode" is a forecast mode that is executed when the weather in the forecast area is likely to be dominated by phenomena that include uncertainties due to the difficulty of predicting cloud areas and cold air masses. Details of the third forecast mode will be described later.
[0054] Figure 6 is a flowchart relating to the first prediction mode (see also Figure 3 as appropriate). The series of processes shown in Figure 6 represent the process of predicting snow depth (S104 in Figure 4) in the first prediction mode (S1032 in Figure 5). As mentioned above, the "first prediction mode" is the prediction mode used when the weather in the target area is likely to be dominated by global phenomena. In this case, high-precision prediction is possible whether one or both GSM and MSM are used. Below, as an example, the case in which both GSM and MSM are used will be explained.
[0055] In step S1041a, the processing unit 12 equalizes the weighting of GSM and MSM. Here, "weighting" refers to the weight coefficient that is multiplied by the calculated snow depth based on the GSM mesh data and the calculated snow depth based on the MSM mesh data when predicting the snow depth in step S1047a, which will be described later. Equalizing the weighting of GSM and MSM means making the aforementioned weight coefficients equal (for example, setting both weight coefficients to 1). In other words, in the first prediction mode, the snow depth is predicted by taking the average of the calculated snow depth based on the GSM mesh data and the calculated snow depth based on the MSM mesh data (S1047a).
[0056] Each of the following steps S1042a to S1047a is performed on a single mesh (the lowest mesh near the ground surface) that includes the area where the solar power generation equipment 40 is located in the GSM or MSM. Alternatively, each of the steps S1042a to S1047a may be performed based on, for example, the analysis value when no artificial errors are introduced (a control run as shown by the thick line in Figure 2A) among multiple ensemble members.
[0057] In step S1042a, the processing unit 12 determines whether the cumulative precipitation is greater than 0. That is, the processing unit 12 determines whether the cumulative precipitation is greater than 0 for a predetermined mesh of the GSM, and also for a predetermined mesh of the MSM. The cumulative precipitation, which is one of the meteorological parameters, is acquired as mesh data in step S101 of Figure 4.
[0058] If the cumulative precipitation is 0 or less in step S1042a (S1042a: No), the processing unit 12 proceeds to step S1043a. In step S1043a, the processing unit 12 determines that the weather in that mesh (i.e., the area where the solar power generation equipment 40 is located) is either sunny or cloudy. In this case, there is little possibility that the amount of power generated by the solar power generation panels will decrease due to the effects of snow. Also, if the cumulative precipitation is greater than 0 in step S1042a (S1042a: Yes), the processing unit 12 proceeds to step S1044a.
[0059] In step S1044a, the processing unit 12 determines that the temperature is a predetermined value TG snow It is less than and the relative humidity is a specified value RH snow Determine whether it is less than or equal to the predetermined value TG. snow or specified value RH snow This is a threshold value that serves as a criterion for determining whether moisture will turn into snow or rain, and it is set in advance.
[0060] In step S1044a, the temperature is a predetermined value TG snow The above, or the relative humidity is at a predetermined value of RH. snow If the above is true (S1044a: No), the processing unit 12 proceeds to step S1045a. In step S1045a, the processing unit 12 predicts that it will rain in that mesh (i.e., moisture will turn into rain). In this case, there is little possibility that the amount of electricity generated by the solar panels will decrease due to the effects of snow.
[0061] Furthermore, in step S1044a, the temperature is a predetermined value TG snow It is less than and the relative humidity is a specified value RH snow If it is less than (S1044a: Yes), the processing unit 12 proceeds to step S1046a. In step S1046a, the processing unit 12 predicts the amount of snowfall based on the cumulative precipitation and the snow-water ratio for the target mesh. Specifically, the processing unit 12 predicts the amount of snowfall by multiplying the cumulative precipitation by the snow-water ratio. Here, the "snow-water ratio" is the ratio of snowfall to 1 mm of precipitation and is calculated based on the temperature near the ground surface. The snow-water ratio may be calculated based on a predetermined formula or data table, or it may be calculated regressively based on past precipitation and snowfall amounts.
[0062] In step S1047a, the processing unit 12 calculates the snow depth (also called snow depth) using the snow depth prediction unit 124. Specifically, the processing unit 12 performs a time integral of the snowfall amount calculated in step S1046a, and calculates the snow depth based on the calculation result and the history information of the snow depth. As described above, the calculation results of the snow depth using GSM and MSM are obtained, and the average value of these is calculated as the prediction result in step S1047a.
[0063] In this way, the snow depth prediction unit 124 predicts the amount of snow in the area where the solar power generation equipment 40 is located by taking the average of the snow depth based on GSM (first mesh data) and the snow depth based on MSM (second mesh data) in the first prediction mode. Then, based on the snow depth predicted in step S1047a, the amount of power generated by the solar power generation equipment 40 is predicted as described above (S107 in Figure 4).
[0064] Figure 7 is a flowchart relating to the second prediction mode (see also Figure 3 as appropriate). The series of processes shown in Figure 7 represent the process of predicting snowfall amount in the second prediction mode (S1034 in Figure 5) (S104 in Figure 4). As mentioned above, the "second prediction mode" is a prediction mode selected when there is a high probability that the weather in the area to be predicted will be dominated by a moderate-scale phenomenon. In this case, geographical factors that are not easily reflected in the GSM (such as coastal topography) greatly influence the presence or absence of snowfall and the amount of snowfall. Therefore, as will be explained below, the weighting of the MSM, which has higher geographical resolution, is made greater than that of the GSM.
[0065] In step S1041b, the processing unit 12 increases the weighting of MSM compared to GSM. As mentioned above, "weighting" refers to the weight coefficient that is multiplied by the calculated snow depth value based on GSM mesh data and the calculated snow depth value based on MSM mesh data in the snow depth prediction (S1047b). For example, the ratio of the weighting of MSM to the weighting of GSM may be set to 2:1, or it may be set to a predetermined ratio such as 5:1, 100:1, or 1:0 (using MSM without specifically using GSM).
[0066] Steps S1042b to S1046b in Figure 7 are the same as steps S1042b to S1046b in Figure 6, so their explanation will be omitted. Then, in step S1047b, the processing unit 12 predicts the amount of snow using the snow depth prediction unit 124. That is, in the second prediction mode, the snow depth prediction unit 124 predicts the amount of snow in the area where the solar power generation facility 40 is located by taking the sum of the value obtained by multiplying the amount of snow based on GSM (first mesh data) by a first weighting coefficient and the value obtained by multiplying the amount of snow based on MSM (second mesh data) by a second weighting coefficient. Here, it is assumed that the second weighting coefficient is larger than the first weighting coefficient. This corresponds to the process of giving more weight to MSM than to GMS (S1041b).
[0067] Figure 8 is a flowchart relating to the third prediction mode (see also Figure 3 as appropriate). The series of processes shown in Figure 8 represent the process of predicting snow depth (S104 in Figure 4) in the third prediction mode (S1035 in Figure 5). As mentioned above, the "third prediction mode" is a prediction mode that is executed when there is uncertainty due to the difficulty of predicting cloud areas and cold air. In this case, the accuracy of normal weather forecasts based on GSM or MSM may be low, so a predetermined machine learning algorithm is used, as will be explained next.
[0068] In steps S1041c to S1048c, MSM mesh data is used as input data for a predetermined prediction model in the machine learning algorithm. The machine learning algorithm uses MSM mesh data with high geographical resolution to perform detailed predictions.
[0069] In step S1041c, the processing unit 12 takes the cumulative precipitation and cloud cover included in the MSM mesh data as input and determines whether or not there is precipitation in the forecast area based on a predetermined machine learning algorithm. Note that the meteorological parameters, including cumulative precipitation and cloud cover, are assumed to have been obtained in step S101 (see Figure 4) (the same applies to the meteorological parameters in S1044c and S1046c described later). In particular, cloud cover has a significant impact on precipitation, so it is desirable to use it as input data for the machine learning algorithm.
[0070] As mentioned above, machine learning algorithms such as neural networks, SVMs (Support-Vector Machines), and decision trees can be used as appropriate (the same applies to S1044c, which will be described later). For example, observational data or numerical weather forecast data may be used as training data to pre-train a predetermined prediction model.
[0071] In step S1042c, the processing unit 12 determines whether or not there is precipitation in the area (a predetermined mesh) where the solar power generation equipment 40 is located. If there is no precipitation in step S1042c (S1042c: No), the processing unit 12 proceeds to step S1043c. In step S1043c, the processing unit 12 determines that the area where the solar power generation equipment 40 is located is either sunny or cloudy.
[0072] Furthermore, in step S1042c, if there is precipitation in the area where the solar power generation equipment 40 is located (S1042c: Yes), the processing unit 12 proceeds to step S1044c. In step S1044c, the processing unit 12 takes the vertical distribution of relative humidity and temperature included in the MSM mesh data as input and determines the presence or absence of snowfall based on a predetermined machine learning algorithm. In particular, the vertical distribution of temperature has a significant impact on snowfall, so it is desirable to use it as input data for the machine learning algorithm.
[0073] In step S1045c, the processing unit 12 determines whether or not there is snowfall in the area (a predetermined mesh) where the solar power generation equipment 40 is located. If there is no snowfall in step S1045c (S1045c: No), the processing unit 12 proceeds to step S1046c. In step S1046c, the processing unit 12 determines that it will rain in the area where the solar power generation equipment 40 is located.
[0074] Furthermore, in step S1045c, if there is snowfall in the area where the solar power generation equipment 40 is located (S1045c: Yes), the processing unit 12 proceeds to step S1047c. In step S1047c, the processing unit 12 takes the cumulative precipitation and vertical temperature distribution included in the MSM mesh data as input and predicts the amount of snowfall based on a machine learning algorithm.
[0075] As mentioned above, machine learning algorithms such as neural networks, SVR (Support-Vector Regression), and multiple regression analysis can be used as appropriate. For example, observational data or numerical weather forecast data may be used as training data to pre-train a predetermined prediction model.
[0076] In step S1048c, the processing unit 12 predicts the amount of snow using the snow depth prediction unit 124. Specifically, the processing unit 12 performs a time integral of the snowfall amount calculated in step S1047c, and predicts the amount of snow in the area where the solar power generation facility 40 is located based on the calculation result and the historical information of the amount of snow.
[0077] Thus, in the third prediction mode, the snow depth prediction unit 124 predicts the amount of snow in the area where the solar power generation facility 40 (see Figure 1) is located, based on the cumulative precipitation, cloud cover, relative humidity, and vertical temperature distribution included in the meteorological parameters of the MSM (second mesh data), using a predetermined machine learning algorithm (S1041c~S1048c). After performing the processing in step S1048c, the processing unit 12 terminates the series of processes related to the third prediction mode (END).
[0078] <Effects> According to the first embodiment, if the variation in the GSM weather parameters is below a first threshold (S1031: Yes in Figure 5), the processing unit 12 selects a first prediction mode (S1032). In the first prediction mode, for example, the weighting of GSM and MSM is set to be equal (S1041a in Figure 6). This makes it possible to predict global phenomena with high accuracy, such as a pressure pattern of high pressure in the west and low pressure in the east caused by the Siberian high.
[0079] Furthermore, if the variation in the GSM weather parameters is higher than the first threshold (S1031: No in Figure 5), and the variation in the MSM weather parameters is below the second threshold (S1033: Yes), the processing unit 12 selects the second prediction mode (S1034). In the second prediction mode, the MSM is weighted more heavily than the GSM (S1041b in Figure 7). By using the MSM, which has a high geographical resolution, medium-scale phenomena such as low-pressure systems along the southern coast can be predicted with high accuracy.
[0080] Furthermore, if the variation in the GSM weather parameters is higher than the first threshold (S1031:No in Figure 5), and the variation in the MSM weather parameters is higher than the second threshold (S1033:No), the processing unit 12 selects the third prediction mode (S1035). In the third prediction mode, snow depth and other factors are predicted based on a machine learning algorithm. This makes it possible to predict phenomena with high accuracy where there is uncertainty due to the difficulty of predicting cloud areas and cold air.
[0081] Thus, according to the first embodiment, it is possible to make highly accurate snowfall predictions while taking into account variations in meteorological parameters (i.e., the risk that predictions from the Japan Meteorological Agency, etc., may be incorrect). This makes it possible to predict the amount of power generated by the solar power generation equipment 40 with high accuracy.
[0082] ≪Second Embodiment≫ The second embodiment differs from the first embodiment in that it corrects the power generation reduction rate based on the error between the predicted and measured snow depth in the solar power generation equipment 40 (see Figure 9). Other aspects are the same as the first embodiment. Therefore, we will explain the differences from the first embodiment, and omit explanations of overlapping parts.
[0083] Figure 9 is a configuration diagram including the solar power generation prediction device 10A according to the second embodiment. The power conditioner 60 shown in Figure 9 converts the power generated by the solar power generation equipment 40 into a predetermined AC power and transmits and receives data related to the solar power generation equipment 40, and is connected to the solar power generation equipment 40 via wiring.
[0084] Furthermore, a snow depth sensor 70 is installed on the solar power generation equipment 40. The snow depth sensor 70 detects the snow depth (depth and amount of snow) on the solar power generation panels. The values detected by the snow depth sensor 70 are transmitted to the solar power generation amount prediction device 10A via the network N1.
[0085] Alternatively, a camera (not shown) may be used instead of the snow depth sensor 70. In this case, the camera will photograph the snow accumulating on the solar power generation panel from the side, and the amount of snow will be calculated based on the captured image. Alternatively, a panel back surface temperature sensor (not shown) that detects the temperature of the back surface of the solar power generation panel may be used instead of the snow depth sensor 70. In this case, the amount of snow on the solar power generation panel will be estimated based on the value detected by the panel back surface temperature sensor, the amount of power generated at any given moment, and the ambient temperature around the solar power generation panel. Note that, assuming that the amount of power generated and the temperature are constant, and that snow is accumulating on the solar power generation panel, there is a tendency for the amount of snow to be greater the lower the temperature of the back surface of the solar power generation panel.
[0086] Figure 10 is a functional block diagram of the solar power generation prediction device 10A. As shown in Figure 10, the processing unit 12A of the solar power generation forecasting device 10A includes a weather parameter acquisition unit 121, a weather parameter variation calculation unit 122, a forecast mode selection unit 123, a snow depth forecast unit 124, a power generation reduction rate calculation unit 125, and a power generation amount forecast unit 126, as well as a power generation reduction rate correction unit 127. In other words, the processing unit 12A is the same as the configuration of the first embodiment (see Figure 3) with the addition of the power generation reduction rate correction unit 127.
[0087] The power generation reduction rate correction unit 127 corrects the power generation reduction rate so that the error between the predicted snow depth in the area where the solar power generation equipment 40 (see Figure 9) is located and the measured snow depth in that area is reduced. Details of the power generation reduction rate correction unit 127 will be described later.
[0088] Figure 11 is a flowchart showing the processes performed by the processing unit of the solar power generation forecasting device (see also Figure 10 as appropriate). Note that steps S101 to S107 in Figure 11 are the same in this order as steps 101 to S107 in the first embodiment (see Figure 4), so their explanation will be omitted. After calculating the amount of power generated in step S107, the processing unit 12A proceeds to step S108. In step S108, the processing unit 12A calculates the error between the predicted snow depth of the solar power generation equipment 40 (calculation result in S104) and the measured snow depth. The target period for acquiring the predicted and measured snow depth is assumed to be the same. The measured snow depth is obtained, for example, from the snow depth sensor 70 (see Figure 9).
[0089] In step S109, the processing unit 12A corrects the power generation reduction rate using the power generation reduction rate correction unit 127 based on the error between the predicted and measured snow depth of the solar power generation equipment 40 (see Figure 9). For example, if the measured snow depth of the solar power generation equipment 40 is greater than the predicted snow depth, there is a high possibility that the power generation reduction rate (calculation result in S106) is underestimated. Therefore, the power generation reduction rate correction unit 127 corrects the power generation reduction rate to be larger.
[0090] Furthermore, if the measured snowfall amount for the solar power generation equipment 40 is smaller than the predicted value, there is a high possibility that the power generation reduction rate is overestimated. In such cases, the power generation reduction rate correction unit 127 corrects the power generation reduction rate to reduce it. It is assumed that the formula or data table used to correct the power generation reduction rate is pre-configured.
[0091] The corrected power generation reduction rate calculated in step S109 is associated with the identification information of the photovoltaic power generation equipment 40 and stored in the memory unit 11 (see Figure 10). The corrected power generation reduction rate is also used to calculate the amount of power generated from the next forecast (the next forecast when the most recent forecast in step S107, based on the power generation reduction rate before correction, is considered the current forecast). After performing the processing in step S109, the processing unit 12A terminates the series of processes (END). Note that the series of processes shown in Figure 11 may be repeated at predetermined intervals.
[0092] <Effects> According to the second embodiment, the power generation reduction rate is appropriately corrected based on the error between the predicted and measured snow depth of the solar power generation equipment 40 (S108, S109 in Figure 11). As a result, the amount of power generated by the solar power generation equipment 40 can be predicted with higher accuracy than in the first embodiment, based on the corrected power generation reduction rate. Therefore, for example, when selling the generated electricity, the accurate value of the amount of power generated can be reflected in the bidding price.
[0093] <<Variations of the second embodiment>> In the second embodiment, a case was described in which the power generation reduction rate correction unit 127 (see Figure 10) corrects the power generation reduction rate based on the error in the amount of snowfall, but it is not limited to this. For example, the power generation reduction rate correction unit 127 may correct the power generation reduction rate so that the error between the predicted value of the amount of power generated by the solar power generation equipment 40 (see Figure 9) and the measured value of the amount of power generated by the solar power generation equipment 40 is reduced.
[0094] For example, if the measured value of the power generation of the solar power generation equipment 40 is greater than the predicted value, it is highly likely that the power generation reduction rate (calculation result in S106) is overestimated. Therefore, the power generation reduction rate correction unit 127 corrects the power generation reduction rate to reduce it. This type of processing also produces the same effect as in the second embodiment. The measured value of the power generation of the solar power generation equipment 40 is transmitted to the solar power generation prediction device 10A sequentially via the power conditioner 60 (see Figure 9) and the network N1 (see Figure 9).
[0095] ≪Third Embodiment≫ The third embodiment differs from the first embodiment in that the solar power generation prediction device 10B (see Figure 12) is equipped with a prediction model localization unit 128 (see Figure 12), which modifies the prediction model to reduce the error between the predicted and measured values of snow depth. Other aspects are the same as the first embodiment. Therefore, the differences from the first embodiment will be explained, and the overlapping parts will be omitted.
[0096] Figure 12 is a functional block diagram of the solar power generation prediction device 10B according to the third embodiment. As shown in Figure 12, the processing unit 12B of the solar power generation forecasting device 10B includes a weather parameter acquisition unit 121, a weather parameter variation calculation unit 122, a forecast mode selection unit 123, a snow depth forecast unit 124, a power generation reduction rate calculation unit 125, and a power generation forecast unit 126, as well as a forecast model localization unit 128. In other words, the processing unit 12B is the same as the configuration of the first embodiment (see Figure 3) with the addition of the forecast model localization unit 128.
[0097] The prediction model localization unit 128 modifies the prediction model of the snow depth prediction unit 124 to a predetermined value (modifying the prediction model to suit the region) so that the error between the predicted snow depth in the region where the solar power generation facility 40 (see Figure 1) is located and the measured snow depth in that region is reduced. Details of the prediction model localization unit 128 will be described later.
[0098] Figure 13 is a flowchart showing the processes performed by the processing unit of the solar power generation forecasting device (see also Figure 12 as appropriate). Note that steps S101 to S108 in Figure 13 are the same in this order as steps 101 to S108 in the second embodiment (see Figure 11), so their explanation will be omitted. After calculating the error between the predicted and measured snow depth of the solar power generation equipment 40 (see Figure 1) in step S108, the processing unit 12B proceeds to step S119.
[0099] In step S119, the processing unit 12B modifies the prediction model of the snow depth prediction unit 124 using the prediction model localization unit 128. That is, the processing unit 12B modifies the prediction model of the snow depth prediction unit 124 so that the error between the predicted value and the measured value of the snow depth is reduced. For example, the snow-water ratio (S1046a in Figure 6 and S1046b in Figure 7), which is one of the prediction models used to calculate the amount of snowfall, may be modified as appropriate. Specifically, if the measured value is greater than the predicted value of the snow depth of the solar power generation equipment 40, the snow-water ratio may be modified so that the amount of snowfall is increased.
[0100] Furthermore, the data table and formulas used to calculate power generation from snow depth, which are part of the prediction model, may be modified as appropriate. In addition, for example, the machine learning algorithms used as prediction models in the third prediction mode (S1041c, S1044c, S1047c in Figure 8) may be modified as appropriate. After performing the processing in step S119, the processing unit 12B terminates the series of processes (END).
[0101] <Effects> According to the third embodiment, the prediction model localization unit 128 appropriately modifies (localizes) the prediction model used to predict snowfall in accordance with the likelihood of snowfall in the area where the solar power generation equipment 40 is installed. This makes it possible to predict the amount of power generated by the solar power generation equipment 40 with higher accuracy than in the first embodiment.
[0102] <<Variation of the third embodiment>> In the third embodiment, a case was described in which the prediction model localization unit 128 (see Figure 12) modifies the prediction model based on the error in the amount of snowfall, but the invention is not limited to this. For example, the prediction model localization unit 128 may modify the prediction model of the snowfall prediction unit 124 so as to reduce the error between the predicted value of the amount of power generated by the solar power generation equipment 40 and the measured value of the amount of power generated by the solar power generation equipment 40. Such processing also produces the same effects as in the third embodiment.
[0103] <<Other variations>> The solar power generation prediction devices 10, 10A, and 10B and the solar power generation prediction method related to this disclosure have been described in each embodiment above. However, this disclosure is not limited to these descriptions, and various modifications can be made. For example, in each embodiment, a case in which the prediction mode is selected based on the variation of the ensemble members (see Figures 2A and 2B) has been described, but this is not limited to this. For example, the prediction mode may be selected based on the variation of a pseudo-ensemble member as shown in Figure 14.
[0104] Figure 14 is an explanatory diagram showing an example of a pseudo-ensemble member acquired in a modified solar power generation prediction device. In Figure 14, the horizontal axis represents time, and the vertical axis represents the values of meteorological parameters. Furthermore, for the forecast area, predetermined meteorological parameters (e.g., temperature, relative humidity, and cumulative precipitation) with different initial times are assumed to be acquired as pseudo-ensemble members. Here, "pseudo-ensemble members" does not mean that artificial errors are introduced into the control run (analyzed values), but rather that they are treated as pseudo-ensemble members.
[0105] In Figure 14, the time tk (k=1,2,···,5) is the initial time (the time corresponding to the initial value) of the k-th pseudo-ensemble member. In this way, the meteorological parameter acquisition unit 121 (see Figure 3) acquires a set of numerical forecasts for a given meteorological parameter, each with a different initial time when the initial value was obtained (5 in the example of Figure 14), as pseudo-ensemble members M1 to M5. The forecast target time for each of the pseudo-ensemble members M1 to M5 is assumed to be time t6.
[0106] The weather parameter variability calculation unit 122 (see Figure 3) calculates the variability of predicted values for weather parameters by treating each of the multiple numerical forecasts as a pseudo-ensemble member. This allows the processing unit 12 (see Figure 3) to calculate the variability of weather parameters even when, for example, the results of the ensemble forecast (i.e., the ensemble members) are unavailable. The estimation of snow depth, etc., may be based on the pseudo-ensemble member M5 whose initial time is closest to the forecast target time (i.e., time t6), or it may be based on the average value of multiple pseudo-ensemble members M1 to M5 at each time.
[0107] Furthermore, in each embodiment, when the variation in the weather parameters of the GSM is below a first threshold (S1031: Yes in Figure 5), the processing unit 12 equalizes the weighting of the GSM and MSM in the first prediction mode (S1032) (S1041a in Figure 6), but the system is not limited to this. For example, in the first prediction mode, the processing unit 12 may use the snow depth prediction unit 124 to predict the amount of snow in the area where the solar power generation facility 40 is located, based on either the GSM (first mesh data) or the MSM (second mesh data). Even with such processing, the amount of snow and the amount of power generated can be predicted with high accuracy.
[0108] Furthermore, although each embodiment describes the use of two types of mesh models (GSM and MSM) with different mesh sizes, there may be three or more types of mesh models. For example, in addition to GSM and MSM, LFM (Local Numerical Weather Prediction Model) may be used. LFM is a mesh model with a horizontal grid spacing of approximately 2 km and 76 layers in the vertical direction, and its mesh is even smaller than that of MSM. For example, if the variation in the meteorological parameters of MSM is greater than the second threshold, and the variation in the meteorological parameters of LFM is less than or equal to the third threshold, the fourth prediction mode is selected. In the fourth prediction mode, the weighting of LFM is set to be greater than that of GSM and MSM. If the variation in the meteorological parameters of LFM is greater than the third threshold, the third prediction mode is selected, and snow depth, etc., is predicted based on a machine learning algorithm.
[0109] When using n different mesh models (where n is a natural number greater than or equal to 2) with varying mesh sizes, the process of comparing the variability of meteorological parameters to a predetermined threshold is performed sequentially, starting with the largest mesh model. If the variability of meteorological parameters in the smallest mesh model exceeds the predetermined threshold, snowfall amount and other parameters are predicted based on a machine learning algorithm.
[0110] Furthermore, the processing performed by the solar power generation prediction device 10 (processing such as the solar power generation prediction method) may be executed as a predetermined program on a computer. The aforementioned program can be provided via a communication line, or it can be written to a recording medium such as a CD-ROM and distributed.
[0111] Furthermore, each embodiment is described in detail for the purpose of clearly illustrating this disclosure and is not necessarily limited to having all the configurations described. Also, it is possible to add, delete, or replace some of the configurations in the embodiments with other configurations. In addition, the mechanisms and configurations described above are those that are considered necessary for explanation and do not necessarily represent all the mechanisms and configurations in the product. [Explanation of symbols]
[0112] 10, 10A, 10B Solar Power Generation Prediction Device 11 Storage section 12, 12A, 12B Processing Unit 13 Input / Output Interfaces 14. Communication Interface 20 Input devices 30 Display device 40 Solar power generation equipment 50 Weather Information Servers 60 Power Conditioner 70 Snow depth sensor 121 Meteorological Parameter Acquisition Unit 122 Meteorological Parameter Variation Calculation Unit 123 Prediction mode selection section 124 Snowfall Forecasting Section 125 Power generation reduction rate calculation unit 126 Power generation forecasting section 127 Power generation reduction rate correction unit 128 Predictive Model Localization Department S101 Step (Meteorological parameter acquisition step) S102 Step (Meteorological parameter variation calculation step) S103 Step (Prediction Mode Selection Step) S104 Step (Snowfall prediction step) S105 Step (Power generation reduction rate calculation step) S107 Step (Power generation prediction step)
Claims
1. The system includes a weather parameter acquisition unit that acquires predicted values for predetermined weather parameters as multiple types of mesh data with different mesh sizes on the map. A solar power generation forecasting device that acquires multiple types of the aforementioned mesh data as a set of multiple numerical forecasts, each with a different magnitude of initial value when predicting the aforementioned weather parameters, A weather parameter variation calculation unit calculates the variation in the predicted values of the weather parameters in the numerical weather forecast for each of the multiple types of mesh data, A prediction mode selection unit selects one of several prediction modes based on the variability of the predicted values of the aforementioned weather parameters. A snow depth prediction unit predicts the amount of snow in an area where solar power generation facilities are located, based on a predetermined prediction mode selected by the prediction mode selection unit, A power generation reduction rate calculation unit calculates the power generation reduction rate of the solar power generation equipment based on the predicted snowfall amount, A solar power generation prediction device further comprising a power generation prediction unit that predicts the amount of power generated by the solar power generation equipment based on the power generation reduction rate.
2. The weather parameter acquisition unit acquires information including, as multiple types of mesh data, first mesh data and second mesh data with a smaller mesh than the first mesh data. The prediction mode selection unit selects the first prediction mode if the variation in the predicted values of the weather parameters in the first mesh data is less than or equal to a first threshold. The snow depth prediction unit predicts the amount of snow in the region by taking the average of the snow depth based on the first mesh data and the snow depth based on the second mesh data in the first prediction mode. A solar power generation prediction device according to claim 1, characterized by the above.
3. The weather parameter acquisition unit acquires information including, as multiple types of mesh data, first mesh data and second mesh data with a smaller mesh than the first mesh data. The prediction mode selection unit selects the first prediction mode if the variation in the predicted values of the weather parameters in the first mesh data is less than or equal to a first threshold. The snow depth prediction unit predicts the amount of snow in the area based on either the first mesh data or the second mesh data in the first prediction mode. A solar power generation prediction device according to claim 1, characterized by the above.
4. The weather parameter acquisition unit acquires information including, as multiple types of mesh data, first mesh data and second mesh data with a smaller mesh than the first mesh data. The prediction mode selection unit selects the second prediction mode if the variation in the predicted values of the weather parameters in the first mesh data is greater than the first threshold, and the variation in the predicted values of the weather parameters in the second mesh data is less than or equal to the second threshold. In the second prediction mode, the snow depth prediction unit predicts the amount of snow in the region by taking the sum of the value obtained by multiplying the snow depth based on the first mesh data by a first weighting coefficient and the value obtained by multiplying the snow depth based on the second mesh data by a second weighting coefficient. The second weighting coefficient is greater than the first weighting coefficient. A solar power generation prediction device according to claim 1, characterized by the above.
5. The weather parameter acquisition unit acquires information including, as multiple types of mesh data, first mesh data and second mesh data with a smaller mesh than the first mesh data. The prediction mode selection unit selects a third prediction mode if the variation in the predicted values of the weather parameters in the first mesh data is greater than a first threshold, and furthermore, the variation in the predicted values of the weather parameters in the second mesh data is greater than a second threshold. In the third prediction mode, the snow depth prediction unit predicts the snow depth of the region using a predetermined machine learning algorithm based on the cumulative precipitation, cloud cover, relative humidity, and vertical temperature distribution included in the meteorological parameters of the second mesh data. A solar power generation prediction device according to claim 1, characterized by the above.
6. The system includes a power generation reduction rate correction unit that corrects the power generation reduction rate so as to reduce the error between the predicted value of the amount of snow in the area where the solar power generation facility is located and the measured value of the amount of snow in the area. A solar power generation prediction device according to claim 1, characterized by the above.
7. The system includes a power generation reduction rate correction unit that corrects the power generation reduction rate so as to reduce the error between the predicted value of the power generation amount of the solar power generation equipment and the measured value of the power generation amount of the solar power generation equipment. A solar power generation prediction device according to claim 1, characterized by the above.
8. The system includes a prediction model localization unit that modifies the prediction model of the snow depth prediction unit so as to reduce the error between the predicted snow depth in the area where the solar power generation facility is located and the measured snow depth in that area. A solar power generation prediction device according to claim 1, characterized by the above.
9. The system includes a prediction model localization unit that modifies the prediction model of the snowfall prediction unit so as to reduce the error between the predicted value of the amount of power generated by the solar power generation equipment and the measured value of the amount of power generated by the solar power generation equipment. A solar power generation prediction device according to claim 1, characterized by the above.
10. The weather parameter acquisition unit acquires a set of multiple numerical weather forecasts for a predetermined weather parameter, each with a different initial time when the initial value was obtained. The weather parameter variation calculation unit calculates the variation in the predicted values of the weather parameters by treating each of the multiple numerical forecasts as a pseudo-ensemble member. A solar power generation prediction device according to claim 1, characterized by the above.
11. The process includes a weather parameter acquisition step that obtains predicted values for predetermined weather parameters as multiple types of mesh data with different mesh sizes on the map. A solar power generation forecasting method, which acquires multiple types of the aforementioned mesh data as a set of multiple numerical forecasts with different initial values for predicting the aforementioned weather parameters, A weather parameter variation calculation step for each of the multiple types of mesh data, which calculates the variation in the predicted values of the weather parameters in the numerical weather forecast, A prediction mode selection step in which one of several prediction modes is selected based on the variability of the predicted values of the aforementioned weather parameters, A snow depth prediction step that predicts the amount of snow in the area where the solar power generation facility is located, based on a predetermined prediction mode selected in the prediction mode selection step, A power generation reduction rate calculation step, which calculates the power generation reduction rate of the solar power generation equipment based on the predicted snowfall amount, A method for predicting solar power generation, further comprising a power generation prediction step of predicting the amount of power generated by the solar power generation equipment based on the power generation reduction rate.