Short-Term Photovoltaic Power Prediction Based on Back Propagation Neural Network Improved by Hybrid Sparrow Algorithm

Accurately predicting short-term photovoltaic (abbr. PV) power generation is the key to increase operational efficiency of PV station and to ensure secure and stable operation of post-grid-connected PV station. Therefore, a back propagation neural network prediction model improved by Elite reverse learning strategy-based hybrid sparrow search algorithm combined with Metropolis criterion was proposed. Firstly, Pearson correlation coefficient formula was utilized to select the meteorological feature set with the best correlation with photovoltaic output as the input of the proposed model to avoid the affection of redundant meteorological factors on photovoltaic output. Secondly, the Euclidean distance formula was utilized to calculate the time series similarity to select the training set to improve the reliability of the training set. Finally, weight threshold value of HSSA-BPNN was used to establish the prediction model, and actual data of a certain PV station located in Xinjiang was used to conduct the experimental analysis. Results of experimental analysis show that the hybrid sparrow search algorithm-BPNN model possesses good adaptability than those from BPNN, particle swarm algorithm-BPNN, and sparrow search algorithm-BPNN, and the prediction performance of the proposed model is better.