Forecasting and uncertainty analysis of day-ahead photovoltaic power using a novel forecasting method

The primary means to promote grid-connected photovoltaic power generation is through accurately forecasting the power output from photovoltaic power stations. This paper proposes a method for day-ahead photovoltaic power forecasting (PPF) and uncertainty analysis using fuzzy c-means (FCM), whale optimization algorithm (WOA), least squares support vector machine (LSSVM), and non-parametric kernel density estimation (NPKDE). The FCM clustering algorithm was used to cluster historical data on numerical weather prediction and photovoltaic power stations, whereby daily data sharing similar meteorological information were clustered into one class. The rapid convergence speed and high convergence accuracy of the WOA were used to optimize the penalty factor and kernel function width of the LSSVM model; this was done to improve the calculation speed and forecasting accuracy of the LSSVM model. The WOA-LSSVM forecasting model was trained using the clustered numerical weather prediction and historical data of a photovoltaic power station. This was subsequently utilized to forecast day-ahead photovoltaic power. The NPKDE method was used to accurately calculate the probability density distribution of forecasting error and the confidence interval of the day-ahead PPF. The root mean square error (RMSE) values of the forecasting power of the WOA-LSSVM, PSO-LSSVM, LSSVM, LSTM and PSO-BP models are 2.55%, 3.00%, 5.60%, 6.03% and 3.18%, respectively, and the calculation results show that the forecasting accuracy of the WOA-LSSVM was higher relative to other models including PSO-LSSVM, LSSVM, LSTM and PSO-BP. Moreover, the NPKDE method was able to accurately describe the probability density distribution of the forecasting error.