Distributed photovoltaic short‐term power forecasting using hybrid competitive particle swarm optimization support vector machines based on spatial correlation analysis

Abstract

Abstract In order to further improve the accuracy of distributed photovoltaic (DPV) power prediction, this paper proposes a support vector machine (SVM) model based on hybrid competitive particle swarm optimization (HCPSO) with consideration of spatial correlation (SC), for realizing short‐term PV power prediction tasks. Firstly, the spatial correlation analysis is conducted on the distributed PV stations. The k‐means clustering method based on morphological similarity distance improvement and mutual information function is used to select the best reference station and the best delay, which generates strongly correlated PV sequences. Then, a hybrid algorithm of particle swarm optimization (PSO) and sine cosine algorithm (SCA) in a competitive framework (HCPSO) is proposed, aiming to fuse the fast convergence capability of PSO algorithm with the global search capability of SCA algorithm, while enabling the algorithm to effectively handle high‐dimensional optimization problems based on a competitive mechanism. Finally, the HCPSO algorithm is combined with SVM algorithm, which expands the applicable scenarios of the SVM model and effectively improves the accuracy of PV short‐term prediction.

Keywords

• distributed power generation
• forecasting theory
• particle swarm optimisation
• renewable energy sources
• support vector machines