Research On Photovoltaic Power Prediction Based On VMD-BRNN-TSP Model

Faced with the depletion of traditional energy sources and increasing environmental pollution,countries around the world are gradually accelerating their researches for alternative energy sources,and the clean,non-polluting and sustainable photovoltaic energy has been widely noticed and studied.In recent years,the global installed capacity of PV system has been increasing,but the problems are also becoming more and more prominent.The process of PV system is influenced by weather and geographical environment,showing volatility and random interference,and its output power is prone to change with external factors,so predicting the power output and reducing the impact of uncertainty are essential to optimize the grid-connected operation of PV system.In this paper,a Time-Sharing prediction model of PV power based on variational mode decomposition(VMD)and Bayesian regularized neural network(BRNN)is proposed.Firstly,the meteorological factor sequence related to the output power is filtered by mutual information analysis.Secondly,the filtered sequence is subjected to VMD to obtain components with different frequencies,which is aimed at reducing the non-stationarity of the data;Then,the similarity and signal-to-noise ratio between the component and the original variable are calculated,after which the key influencing factors of the original variable are screened out to eliminate the correlation and redundancy of the data.Finally,divide the reconstructed sequence into day and night based on whether the irradiance is 0.Meanwhile,BRNN is used to model the 2datasets respectively and then the prediction results are arranged in time series.It was verified by experimentally analyzed that the mean absolute error(MAE)of the method proposed in this paper is 0.1281,which is reduced by 40.28%compared with the single BP neural network model on the same dataset,and the mean absolute percentage error(MAPE)is 0.0033,the coefficient of determination(R~2)is 0.9907,and other error indicators also confirm that variational modal decomposition is meaningful and Time-Sharing prediction is of much superiority.