Short-term Photovoltaic Output Forecasting with Weakly Related Meteorological Data

Photovoltaic (PV) output is influenced by meteorological factors, and the significant degree of meteorological data influences the accuracy of forecasting result. In this paper, a short-term PV output forecasting method is presented in such weather situation that the weather data and PV output data are weakly correlated. The main factors affecting PV generation are found out by Pearson correlation coefficient method. Based on relevant factors, fuzzy clustering analysis method is used to build similar days, and support vector regression (SVR) forecasting model that has excellent learning ability for small sample is built. In order to determine model parameters, a two-step parameter-determining method is proposed, in which the global grid searching method is applied to determine the value region of kernel parameter p and regularization parameter C which are optimized through self-adaptive differential evolution algorithm, then the prediction accuracy is increased when the appropriate ranges of parameter ε are wider. Examples show that the proposed SVR method has good forecasting ability when the weather data and PV output data are weakly correlated, and the average prediction value, RMSE, is 5.551% which meets the requirement of prediction, and the accuracy is increased to 1.238% by comparison with general BP forecasting method.