Temporal And Spatial Distribution Characteristics And Prediction Of Solar Irradiance In Anhui Province

As the amount of installed photovoltaic(PV)continues to increase,large-scale PV is integrated into the grid.Since PV power generation is intermittent and fluctuating,it will have a greater impact on PV grid connection and the safe and stable operation of the power system.Solar irradiance as an important factor affecting PV power generation,accurate prediction of solar irradiance is important for PV power generation,which is conducive to power system regulation and planning of power resources,improving resource utilization and ensuring safe operation of power grid.Solar irradiance is affected by various factors,which is difficult to be predicted accurately by traditional methods.With the application of neural network models in the field of solar irradiance prediction,the prediction effect has been greatly improved.In this thesis,spatial and temporal distribution analysis methods and methods based on wavelet packet decomposition and Bi-directional Gated Recurrent Unit(Bi-GRU)neural network are used to analyze and predict the distribution of solar irradiance The details of the study are as follows.(1)The solar irradiance data of Anhui Province from 2014 to 2018 were selected for spatial and temporal analysis.It was found that the solar irradiance in Anhui Province as a whole showed a zonal distribution according to latitude.The solar irradiance in Huaibei Plain showed a high value aggregation state,and the hilly and mountainous areas in western Anhui showed a low value aggregation state.Solar irradiance is highest in July and lowest in December.The variation of solar irradiance in Anhui from 2014 to 2018 was significantly different,which was related to solar activity and weather anomalies.The solar irradiance in 2014 was the highest in five years and the lowest in 2016.(2)Multiple correlation analysis was performed between the raw solar irradiance data and eight meteorological factors,including barometric pressure,air temperature,precipitation,evaporation,relative humidity,wind speed,sunshine hours and 0 cm ground temperature,to quantify the degree of influence of each factor on the solar irradiance decomposition coefficient,and monthly forecasts were made by season using a Bi-directional Gated Recurrent Unit(Bi-GRU)and compared with Support Vector Machine(SVM)and Autoregressive Integrated Moving Average Model(ARIMA)for comparison.(3)The wavelet packet decomposed solar irradiance data were normalized and predicted by Bi-GRU neural network and then reconstructed,and the results were compared with those predicted directly by Bi-GRU neural network.Finally,the prediction results are analyzed by spatial interpolation.The results showed that the model predicted by wavelet packet decomposition exhibited smaller prediction errors in all seasons.In the spring forecast,the mean absolute error(MAE)decreased by 20.1%,the root mean square error(RMSE)decreased by about 24.0%,and the R~2 score improved from 0.55 to 0.74.In the summer forecast,the MAE decreased by 44.1%,the RMSE decreased by about 36.9%,and the R~2 improved from-0.30 to 0.48.In the fall forecast,the MAE decreased by 13.6%,the RMSE decreased by 13.5%,and the R~2score improved from 0.15 to 0.37.In the winter forecast,MAE decreased by 8.65%,RMSE decreased by 9.81%,and the R~2 score improved from-0.9 to-0.54,the prediction results are reliable.