A solar radiation forecasting system and method based on multi-source algorithm integration

Abstract

The present invention discloses a solar radiation forecasting system and method integrated with a multi-source algorithm, comprising: obtaining data 1 and data 2 based on radiation forecast data; interpolating data 2 onto hourly data to obtain data 3, interpolating the two sets of data onto a number of radiation stations, combining radiation hourly observation data, obtaining two sets of data station weight coefficient tables through data processing, flattening the two sets of data station weight coefficient tables onto a grid grid field consistent with the spatial resolution of the two sets of data sources, obtaining a gridded weight coefficient table, wherein the two sets of data are data 1 and data 3; integrating the two sets of data based on the two sets of data grid weight coefficient tables to obtain final radiation data. The present invention is applicable to the intersection of energy and meteorological forecasting, can avoid the singleness of the data source and the limitations in certain areas, improve the forecasting effect, and can also improve the quality of radiation forecasting, and enhance the accuracy and stability of forecasting results.

Description

Technical Field

[0001] The present invention belongs to the intersection field of energy and meteorological forecasting, and specifically relates to a solar radiation forecasting system and method integrating multi-source algorithms. Background Art

[0002] Solar radiation is closely related to our lives. It refers to the total short-wave solar radiation received by the ground. It is an important indicator of a place’s solar energy resources and the main factor affecting the power supply of solar power stations.

[0003] However, the existing solar radiation forecasting method adopts the numerical forecasting model. Although it can better predict the solar radiation reaching the ground and output gridded forecast products, the single data source has certain limitations for forecasts in certain areas, making it difficult to make accurate predictions and the forecasting effect is poor. At the same time, the data of the existing method is single and unstable, which makes the accuracy of the forecast results not high. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the prior art and provide a solar radiation forecasting system and method integrating a multi-source algorithm.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A solar radiation forecasting method integrating multi-source algorithms comprises the following steps:

[0007] Obtain data 1 and data 2, interpolate data 2 to hourly data to obtain data 3, interpolate the two sets of data to several radiation stations, and obtain the forecast data set participating in the training, wherein data 1 is the regional model radiation forecast data with a resolution of 9 km and an hourly time resolution, and data 2 is the global model radiation forecast data with a resolution of 9 km and an accumulated time resolution of 3 hours. The two sets of data are data 1 and data 3;

[0008] The two sets of data are processed to obtain two sets of data site weight coefficient tables;

[0009] The two sets of data station weight coefficient tables are flattened onto a grid field that is consistent with the spatial resolution of the two sets of data sources, and the weight coefficient on each grid is calculated to obtain a gridded weight coefficient table;

[0010] Based on the gridded weight coefficient table, the radiation forecast grid data are integrated to obtain the gridded hourly radiation integrated forecast product.

[0011] Preferably, the acquisition of data 1 and data 2 includes: selecting a training period of a certain length, screening the national radiation observation stations, eliminating abnormal stations and stations with too few samples, selecting several stations to participate in training, and forming radiation observation data participating in the training; obtaining radiation forecast data 1 and data 2 for the same training period.

[0012] Preferably, the data processing includes:

[0013] Based on hourly radiation observation data, the forecast quality of the two sets of data is verified, and the root mean square error of the station forecast data based on the two sets of data is solved;

[0014] The root mean square error obtained from the four seasons and three-day forecast time is averaged seasonally and averaged over the forecast time to obtain the root mean square error of all radiation stations under the four seasons and three-day forecast time. The weight coefficient table of the two sets of data stations under the four seasons and three-day forecast time is solved using the weight calculation formula, wherein the weight calculation formula is:

[0015]

[0016] Where k = 1 or 2, RMS E(1, j) and RMS E(2, j) are the root mean square errors of the two sets of data at the jth station, W 1，j and W 2,j is the weight of the two sets of data at the jth site.

[0017] Preferably, the radiation forecast grid data is integrated, and the calculation formula is:

[0018] Y(i,j)=Y1(i,j)×a1(i,j)+Y2(i,j)×a2(i,j),

[0019] Among them, Y(i,j) is the radiation integrated forecast data, Y1(i,j) and Y2(i,j) are two sets of data located at the longitude and latitude grid points (i,j), a1(i,j) and a2(i,j) are the weights of the two sets of data at the same point, and the sum of a1(i,j) and a2(i,j) is 1.

[0020] A solar radiation forecasting system integrating multi-source algorithms, including:

[0021] Data pre-processing module: used to obtain data 1 and data 2, interpolate data 2 to the hourly data to obtain data 3, and interpolate the two sets of data to several radiation stations to obtain the training data set;

[0022] Weight coefficient calculation module: used to process two sets of data to obtain two sets of data site weight coefficient tables;

[0023] Weight coefficient gridding module: used to flatten the weight coefficient tables of the two data stations to a grid field consistent with the spatial resolution of the two data sources, calculate the weight coefficient on each grid, and obtain the gridded weight coefficient table;

[0024] Real-time radiation integrated forecast module: used to integrate the radiation forecast grid data based on the grid weight coefficient table to obtain the gridded hourly radiation integrated forecast product.

[0025] Preferably, the data pre-processing module includes a historical forecast data acquisition unit, a data decoding and extraction unit, a data interpolation unit, an observation data acquisition unit, a real-time data acquisition unit, a data quality control unit, a data pre-processing unit and a sample matching unit.

[0026] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0027] The present invention interpolates two sets of data to several radiation stations and uses a weight calculation formula to solve the two sets of data source station weight coefficient tables under four seasons and three-day forecast time, thereby avoiding the singleness of the data source and the limitations in certain areas and improving the forecast effect; at the same time, the present invention integrates the two sets of data based on the two sets of data grid weight coefficient tables to obtain the final radiation data, effectively improving the quality of radiation forecasting, and enhancing the accuracy and stability of the forecast results. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a flow chart of a solar radiation forecasting system and method integrated with a multi-source algorithm of the present invention;

[0029] Figure 2 It is a system structure diagram of a solar radiation forecasting system and method integrated with a multi-source algorithm of the present invention;

[0030] Figure 3 The present invention discloses a system workflow diagram of a solar radiation forecasting system and method integrating multiple source algorithms. DETAILED DESCRIPTION

[0031] The following is combined with Figure 1-3 , further illustrating a specific implementation of a solar radiation forecasting system and method integrating a multi-source algorithm of the present invention. A solar radiation forecasting system and method integrating a multi-source algorithm of the present invention is not limited to the description of the following embodiments.

[0032] Embodiment 1:

[0033] This embodiment provides a specific structure of a solar radiation forecasting method integrating a multi-source algorithm, such as Figure 1 As shown, the following steps are included:

[0034] Get data 1 and data 2, interpolate data 2 to hourly data, get data 3, interpolate the two sets of data to several radiation stations, and get the forecast data set for training, where data 1 is the regional model radiation forecast data with a resolution of 9 km and an hourly time resolution, and data 2 is the global model radiation forecast data with a resolution of 9 km and an accumulated time resolution of 3 hours. The two sets of data are data 1 and data 3;

[0035] The two sets of data are processed to obtain two sets of data site weight coefficient tables;

[0036] The two sets of data station weight coefficient tables are flattened onto a grid field that is consistent with the spatial resolution of the two sets of data sources, and the weight coefficient on each grid is calculated to obtain a gridded weight coefficient table;

[0037] Based on the gridded weight coefficient table, the radiation forecast grid data are integrated to obtain the gridded hourly radiation integrated forecast product.

[0038] Specifically, obtaining data 1 and data 2 includes: selecting a training period of a certain length, screening the national radiation observation stations, eliminating abnormal stations and stations with too few samples, selecting several stations to participate in training, and forming radiation observation data participating in the training; obtaining radiation forecast data 1 and data 2 for the same training period.

[0039] Specifically, data processing includes:

[0040] Based on hourly radiation observation data, the forecast quality of the two sets of data is verified, and the root mean square error of the station forecast data based on the two sets of data is solved;

[0041] According to the root mean square error obtained in four seasons and three-day forecast time, the seasonal average and forecast time average are performed to obtain the root mean square error of all radiation stations under four seasons and three-day forecast time. The weight calculation formula is used to solve the weight coefficient table of two sets of data stations under four seasons and three-day forecast time. The weight calculation formula is:

[0042]

[0043] Where k = 1 or 2, RMS E(1, j) and RMS E(2, j) are the root mean square errors of the two sets of data at the jth station, W 1，j and W 2,j is the weight of the two sets of data at the jth site.

[0044] Furthermore, the radiation forecast grid data is integrated and the calculation formula is:

[0045] Y(i,j)=Y1(i,j)×a1(i,j)+Y2(i,j)×a2(i,j),

[0046] Among them, Y(i,j) is the radiation integrated forecast data, Y1(i,j) and Y2(i,j) are two sets of data located at the longitude and latitude grid points (i,j), a1(i,j) and a2(i,j) are the weights of the two sets of data at the same point, and the sum of a1(i,j) and a2(i,j) is 1.

[0047] A system for realizing a solar radiation forecasting method integrating multi-source algorithms, such as Figure 2 and Figure 3 As shown, including:

[0048] Data pre-processing module: used to obtain data 1 and data 2, interpolate data 2 to the hourly data to obtain data 3, and interpolate the two sets of data to several radiation stations to obtain the training data set;

[0049] Weight coefficient calculation module: used to process two sets of data to obtain two sets of data site weight coefficient tables;

[0050] Weight coefficient gridding module: used to flatten the weight coefficient tables of the two data stations to a grid field consistent with the spatial resolution of the two data sources, calculate the weight coefficient on each grid, and obtain the gridded weight coefficient table;

[0051] Real-time radiation integrated forecast module: used to integrate the radiation forecast grid data based on the grid weight coefficient table to obtain the gridded hourly radiation integrated forecast product.

[0052] Specifically, the data pre-processing module includes a historical forecast data acquisition unit, a data decoding and extraction unit, a data interpolation unit, an observation data acquisition unit, a real-time data acquisition unit, a data quality control unit, a data pre-processing unit and a sample matching unit.

[0053] Embodiment 2:

[0054] This embodiment provides a specific structure of a solar radiation forecasting system and method integrating a multi-source algorithm, such as Figure 1 As shown, taking the national solar radiation pattern data and observation data for the four seasons from September 1, 2016 to August 31, 2017 as an example, the equal-weight integration method and the integration method based on the weight coefficient table are used to integrate the data based on data 3 and data 1, respectively, and the forecast data based on the weight coefficient table integration method, the forecast data of the equal-weight integration method and the original forecast data are compared and tested and analyzed, which further illustrates the advantages of the solar radiation forecasting system and method integrated with a multi-source algorithm of the present invention.

[0055] First, based on the hourly radiation observation data, Data 1 and Data 2 are obtained, where Data 1 is the regional model radiation forecast data with a resolution of 9 km and an hourly time resolution, and Data 2 is the global model radiation forecast data with a resolution of 9 km and an accumulated time resolution of 3 hours.

[0056] Data 2 is interpolated to the hourly data to obtain data 3, and the two sets of data are interpolated to several radiation stations, where the two sets of data are data 1 and data 3;

[0057] The average root mean square error of the two sets of data at the national radiation stations in different seasons and different forecast time periods was calculated. The root mean square error obtained in four seasons and three-day forecast time periods was averaged seasonally and averaged for the forecast time period to obtain the root mean square error of all radiation stations averaged over the four seasons and three-day forecast time period. The weight calculation formula was used to solve the weight coefficient table of the two sets of data source stations under the four seasons and three-day forecast time period, where the weight calculation formula is:

[0058]

[0059] Where k = 1 or 2, RMS E(1, j) and RMS E(2, j) are the root mean square errors of the two sets of data at the jth station, and W1, j and W2, j are the weights of the two sets of data at the jth station;

[0060] Any point in the radiation grid field is a spatial coordinate point with longitude and latitude. The discrete point data gridding method is used to interpolate the station weight coefficient table to a 9km grid field consistent with the spatial resolution of the two sets of data. Two sets of data grid weight coefficient tables for four seasons and three-day forecast time are obtained, totaling 12 sheets.

[0061] Based on the terrain, 12 grid point weight coefficient tables are determined for the four seasons of spring, summer, autumn and winter and the first, second and third days of the forecast. The weights do not change when the longitude and latitude of the grid points and the longitude and latitude of the radiation stations remain unchanged.

[0062] Secondly, based on the two sets of data grid weight coefficient tables, the two sets of data are integrated according to the weight coefficient table integration method to obtain the forecast data according to the weight coefficient table integration method. The calculation formula is:

[0063] Y(i,j)=Y1(i,j)×a1(i,j)+Y2(i,j)×a2(i,j),

[0064] Among them, Y(i,j) is the forecast data based on the weight coefficient table integration method, Y1(i,j) and Y2(i,j) are two sets of data located at the longitude and latitude grid points (i,j), a1(i,j) and a2(i,j) are the weights of the two sets of data at the same point, and the sum of a1(i,j) and a2(i,j) is 1;

[0065] Then, data 3 and data 1 are integrated according to a 1:1 weight ratio to obtain the forecast data of the equal weight integration method;

[0066] Finally, the forecast data based on the weight coefficient table integration method, the forecast data based on the equal weight integration method and the original forecast data are compared and tested:

[0067] The national radiation stations are matched one by one with the two sets of forecast data grid points through bilinear interpolation. In the process of data processing, abnormal data are eliminated and quality control is carried out. In the process of inspection, indicators such as data absolute error AE, mean absolute error MAE, correlation coefficient CO and root mean square error RMSE are used to characterize the forecast quality of numerical forecast data. The forecast effects of the two forecast models are analyzed. The calculation formulas of each forecast indicator are as follows:

[0068] Absolute error: AE j =|f j -o j |

[0069] Mean absolute error:

[0070] Root Mean Square Error:

[0071] Correlation coefficient:

[0072] Among them, f and o are the radiation values ​​of the forecast field and the actual field, and is the average value, N is the number of samples, and i is the sample number of the forecast and observation.

[0073] Data 1 and data 2 with a time resolution of 1h obtained by time interpolation are interpolated to radiation stations across the country using spatial interpolation. According to the spatial distribution of the correlation coefficient and root mean square error of data 1 and data 2 interpolated to the radiation station, it can be clearly seen that the correlation of the two models shows that the correlation coefficient in Northeast my country and North China is larger, the root mean square error is smaller, and the forecast effect is better, followed by Central China and South China. The correlation coefficient in Northwest China is smaller, the root mean square error is larger, the correlation coefficient between the Qinghai-Tibet Plateau and Xinjiang is the smallest, the correlation is the weakest, and the forecast effect is the worst. The best forecast data in different regions are different. Therefore, considering the excellence of the two sets of data in different forecast effects, an integrated solution should be adopted.

[0074]

[0075] Table 1

[0076] As shown in Table 1, the overall average correlation coefficient CO, root mean square error RMSE and mean absolute error MAE of different radiation integration schemes under different forecast time periods in spring, summer, autumn and winter, A, B, C and D in the table represent data 2, data 1, the integration scheme of two data according to the weight coefficient table, and the integration scheme of two data according to 1:1 equal weight respectively;

[0077] By comparing the forecast data of the weight coefficient table integration method, the forecast data of the equal weight integration method and the original forecast data, we found that from the perspective of correlation:

[0078] The forecast data of the integration method based on the weight coefficient table > the forecast data of the equal weight integration method > data 2 > data 1;

[0079] From the perspective of root mean square error and mean absolute error:

[0080] The forecast data of the integration method based on the weight coefficient table < the forecast data of the equal weight integration method < data 2 < data 1;

[0081] Conclusion: The forecast effect of the data obtained by integrating the weight coefficient is the best, and the forecast effect of the data integrated by equal weight is second, but both are better than the original two sets of data without integration, among which the forecast effect of the original 9km regional model data is the worst.

[0082] Comparing the test results of different seasons, it is found that in winter, the correlation between the forecast data and the observed data is good, with a correlation coefficient of up to 0.88, and the root mean square error and absolute error of the forecast data are the smallest, as low as 116 (wm-2) and 101 (wm-2), respectively. That is, the radiation data forecast effect is best in winter, followed by spring and autumn, and the worst in summer.

[0083] By comparing the test results of different forecast time periods, it was found that the forecast effects on the first, second and third days were not much different, but overall the forecast effect on the first day was better, followed by the second day, and the forecast effect on the third day was relatively poor.

[0084] Working principle: Figure 1-3 As shown, the working process of the data pre-processing module is as follows:

[0085] First, the historical forecast data is collected by the historical forecast data collection unit to obtain data 1 and data 2, and then the historical forecast data is processed by the data decoding and extraction unit. The processed data is interpolated by the data interpolation unit to obtain data 3, which is input into the sample matching unit;

[0086] Then, the observation data is collected by the observation data collection unit, and then the observation data is processed by the data quality control unit and input into the sample matching unit; the sample matching unit inputs the two sets of data into the weight coefficient calculation module;

[0087] Finally, the real-time data is collected by the real-time data collection unit, and then processed by the data pre-processing unit and input into the real-time radiation integrated forecast module;

[0088] Working process of weight coefficient calculation module: Based on hourly radiation observation data, check the forecast quality of two sets of data, and solve the root mean square error of station forecast data based on the two sets of data; perform seasonal average and forecast time average according to the root mean square error obtained in four seasons and three days of forecast time, and obtain the root mean square error of all radiation stations under four seasons and three days of forecast time, and use the weight calculation formula to solve the weight coefficient table of two sets of data stations under four seasons and three days of forecast time;

[0089] The working process of the weight coefficient gridding module is as follows: the weight coefficient tables of the two data stations are flattened to a grid field with the same spatial resolution as the two data sources, the weight coefficient on each grid is calculated, and the gridded weight coefficient table is obtained;

[0090] Working process of real-time radiation integrated forecast module: Based on the gridded weight coefficient table, the radiation forecast grid data is integrated to obtain the gridded hourly radiation integrated forecast product.

[0091] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A solar radiation forecasting method integrating multi-source algorithms, characterized in that , including the following steps: Obtain data 1 and data 2, interpolate data 2 to hourly data to obtain data 3, interpolate the two sets of data to several radiation stations, and obtain the forecast data set participating in the training, wherein data 1 is the regional model radiation forecast data with a resolution of 9 km and an hourly time resolution, and data 2 is the global model radiation forecast data with a resolution of 9 km and an accumulated time resolution of 3 hours. The two sets of data are data 1 and data 3; The two sets of data are processed to obtain two sets of data site weight coefficient tables; The data processing includes: Based on hourly radiation observation data, the forecast quality of the two sets of data is verified, and the root mean square error of the station forecast data based on the two sets of data is solved; The root mean square error obtained from the four seasons and three-day forecast time is averaged seasonally and averaged over the forecast time to obtain the root mean square error of all radiation stations under the four seasons and three-day forecast time. The weight coefficient table of the two sets of data stations under the four seasons and three-day forecast time is solved using the weight calculation formula, wherein the weight calculation formula is: ， where k = 1 or 2, and is the root mean square error of the two sets of data at the jth station, W 1，j and W 2,j is the weight of the two sets of data at the jth site; The two sets of data station weight coefficient tables are flattened onto a grid field that is consistent with the spatial resolution of the two sets of data sources, and the weight coefficient on each grid is calculated to obtain a gridded weight coefficient table; Based on the gridded weight coefficient table, the radiation forecast grid data are integrated to obtain the gridded hourly radiation integrated forecast product.

2. A solar radiation forecasting method integrating multi-source algorithms as claimed in claim 1, characterized in that: The obtaining of data 1 and data 2 includes: selecting a training period of a certain length, screening the national radiation observation stations, eliminating abnormal stations and stations with too few samples, selecting several stations to participate in training, and forming radiation observation data participating in the training; obtaining radiation forecast data 1 and data 2 for the same training period.

3. The solar radiation forecasting method integrating multiple source algorithms as claimed in claim 1, characterized in that: The radiation forecast grid data is integrated, and the calculation formula is: Y(i,j)=Y1(i,j)×a1(i,j)+Y2(i,j)×a2(i,j), Among them, Y(i,j) is the radiation integrated forecast data, Y1(i,j) and Y2(i,j) are two sets of data located at the longitude and latitude grid points (i,j), a1(i,j) and a2(i,j) are the weights of the two sets of data at the same point, and the sum of a1(i,j) and a2(i,j) is 1.

4. A solar radiation forecasting system integrating a multi-source algorithm according to any one of claims 1 to 3, characterized in that: include: Data pre-processing module: used to obtain data 1 and data 2, interpolate data 2 to the hourly data to obtain data 3, and interpolate the two sets of data to several radiation stations to obtain the training data set; Weight coefficient calculation module: used to process two sets of data to obtain two sets of data site weight coefficient tables; Weight coefficient gridding module: used to flatten the weight coefficient tables of the two data stations to a grid field consistent with the spatial resolution of the two data sources, calculate the weight coefficient on each grid, and obtain the gridded weight coefficient table; Real-time radiation integrated forecast module: used to integrate the radiation forecast grid data based on the grid weight coefficient table to obtain the gridded hourly radiation integrated forecast product.

5. A solar radiation forecasting system integrating multiple source algorithms as claimed in claim 4, characterized in that: The data pre-processing module includes a historical forecast data acquisition unit, a data decoding and extraction unit, a data interpolation unit, an observation data acquisition unit, a real-time data acquisition unit, a data quality control unit, a data pre-processing unit and a sample matching unit.

Images