An experimental investigation of two Wavelet-MLP hybrid frameworks for wind speed prediction using GA and PSO optimization

Abstract

Wind speed series show unsteady and nonlinear phenomena. The accurate forecast of wind speed is important for the safety of renewable energy utilization. Compared to the prediction models which use single algorithms, hybrid models always have higher accuracy. Based on the theories of Wavelet, classical time series analysis, genetic algorithm, particle swarm optimization and artificial neural networks, two hybrid forecasting frameworks [the Wavelet-Genetic Algorithm (GA)-Multilayer Perceptron (MLP) and the Wavelet-Particle Swarm Optimization (PSO)-Multilayer Perceptron (MLP)] are proposed to predict non-stationary wind speeds. Comparisons of forecasting performance using different algorithm combinations are provided to investigate the contribution of different components in those two hybrid frameworks. The results based on three experimental cases show that: (1) both of the two proposed hybrid forecasting frameworks are suitable for the diverse accuracy requirements in wind speed predictions, which can be applied to wind power systems; and (2) in both of the two hybrid frameworks, the contribution of the GA and the PSO components in improving the MLP are not statistically significant while that of the Wavelet component is statistically significant.