Evaluation And Improvement Of Solar Radiation Prediction Methods Based On Weather Forecast Informatio

Accurate prediction of daily solar radiation（R_s）is crucial for photovoltaic power generation and many other industries.Reliable observation of solar radiation is a global challenge.Using mechanism models combined with public weather forecast information is a feasible method for predicting short-term solar radiation.This article aims to explore using only public weather forecast data（temperature,weather type）to improve the prediction accuracy of sunshine hours and thereby improve the accuracy of solar radiation prediction.Therefore,this paper proposes a sunshine hour conversion method(S_new)based on predicted temperature and weather type data,and validates it using measured data from 86 radiation observation stations.Finally,the optimal solar radiation model based on sunshine hours（R_s＿S）is combined with the S_newmethod（R_s＿new）to predict solar radiation,and compared with using only weather type data(R_s＿S_com)and a temperature based solar radiation model（R_s＿T）.The conclusions are as follows:（1）Based on the observation data of 96 radiation observation stations from 1967 to 2016,the accuracy of 32 models（S1 to S32）based on sunshine hours was evaluated.The results show that the performance of S18 to S32 models is good and close,and the average values of their regression coefficient（b）,determination coefficient（R²）,root mean square error（RMSE）,and model efficiency（EF）for a single site are 0.975 to 0.976,0.778 to 0.785,2.368to 2.412 MJ/（m²·d）,and 0.881 to 0.885,respectively.After further generalizing the S18 to S32 models,it was found that the S18 model only requires three parameters,with high estimation accuracy and good stability（ranking second overall）.It is recommended that this model be a general model for estimating daily solar radiation in China’s non radiation observation areas.（2）Compared with the traditional sunshine hour estimation method(S_com),the S_newmethod has improved the R²and consistency coefficient(d _IA)of sunshine hour（S）prediction accuracy by 27.8%to 55.7%and 2.9%to 9.5%,respectively,in the 1-7 day forecast at 86stations,while the RMSE has decreased by 12.8%to 14.8%.（3）Prediction of 1～7 day solar radiation based on data from 86 radiation observation stations:R_s＿S_newmodel compared to R_s＿S_commodel has higher accuracy.In the 1～7 day forecast,the mean values of R²and d _IAincreased by 16.8%to 23.1%and 2.1%to 4.8%respectively,while RMSE decreased by 9.7%to 12.5%.R_s＿S_newmodel is dominant in predicting short-term solar radiation,with R_s＿S_newIn the 1～7 day forecast of the S_newmodel,the proportion of global performance index（GPI）ranking first among 86 stations is 52.3%to74.4%,followed by R_s＿T model（25.6%～47.7%）.In addition,R_s＿S_newperformance of the S_newmodel in predicting daily solar radiation is better than other models as the prediction time increases.Therefore,it is recommended to use R_s＿S_newforecasts short-term daily solar radiation.