Photovoltaic Power Prediction Based On Machine Learning

Due to environmental degradation and energy shortages,new energy sources,mainly clean and renewable energy sources have attracted attention.Solar energy is not only huge,rich in resources,but also clean and pollution-free,consequently,photovoltaic(PV)power generation has been rapidly developed and applied.With the continuous increase of PV installed capacity,the impact of its random and fluctuating power generation characteristics on the power system cannot be ignored.The forecast of PV power generation in this paper can provide decision-making reference for dispatchers to formulate scheduling plans,reduce the adverse impact of PV access on the power grid,and improve the safety and stability of system operation;On the other hand,that helps improve the absorptive capacity of PV energy,obtain greater economic and social benefits.This paper firstly conducts qualitative and quantitative analysis of the influencing factors of PV power generation,and screens for the main influencing factors of PV output.Focus on preprocessing the original data set,including normalization,removal and addition of abnormal data,data classification,and dimensionality reduction processing to ensure the accuracy of the training data.Next,a short-term forecast of PV power based on the back propagation neural network(BP network)is established.Through simulation examples,it is found that the modeling process of the BP network is complex and easily falls into local minimums.In view of the shortcomings of the BP network model,this paper uses Support Vector Machine(SVM)regression algorithm to predict the PV output point,and combines the advantages of Particle Swarm Optimization(PSO)and Grid Search Method(GSM)to optimize SVM parameters(c,g),proposes a PSO-GSM-SVM-based PV power short-term point prediction model.Randomly predict the PV power of a certain day in clear and non-clear sky conditions,and compare it with other point prediction models.The results show that the PSO-GSM-SVM-based PV power short-term point prediction model can adapt to different weather conditions and the model has strong generalization ability,high fitting degree and fast running speed.Aiming at the point prediction model cannot quantify the error impact caused by PV volatility and randomness,the concept and application of the forecasting interval are introduced,and a short-term interval forecasting model of PV power based on the lower upper bound estimation theory(LUBE)is proposed.An optimization objective model based on prediction interval accumulative deviation(PIAD)was introduced,and the weight of LUBE was optimized using the PSO algorithm.The PSO-LUBE short-term photovoltaic power interval prediction model based on the optimization objective of PIAD was constructed.Simulation results show that the model meets the confidence requirements under different weather conditions to ensure the reliability of the prediction interval.At the same time,the obtained prediction interval normalized average width is narrow and the PIAD is small,which ensures the accuracy of the prediction interval.