A Method to Forecast Ultra-Short-Term Output of Photovoltaic Power Generation Based on Chaotic Characteristic-Improved Whale Optimization Algorithm and Relevance Vector Machine

By means of mining the chaotic characteristic of photovoltaic(PV) sequence, a forecasting method based on chaotic characteristic-improved whale optimization algorithm (abbr. WOA) and relevance vector machine (abbr. RVM) was proposed to determine physical relation between RVM parameters and chaotic characteristic of the output of PV power generation. Firstly, the chaotic parameters of PV sequence were computed by pseudo nearest neighbor method and complex autocorrelation method to reconstruct the phase space and to determine the radial range of the Gaussian kernel by RVM prediction method. Secondly, the optimization of RVM parameters was implemented by WOA to improve the generalization ability and the convergence rate of the kernel function to perfect the ultra-short-term forecasting method. Finally, by use of output data of practical PV power generation, the forecasted effect of five typical days was simulated and analyzed. Simulation results show that under different weather situations, the proposed method possesses both satisfied forecasting accuracy and adaptability.