Evaluating a dendritic neuron model for wind speed forecasting

Abstract

Because of the intrinsic complexity and chaotic nature of wind speed time series, an appropriate model for accurately forecasting the moving tendency is required. In this paper, we propose an evolutionary dendritic neuron model (EDNM) to carry out wind speed forecasting. The model is trained via adaptive differential evolution with the linear population size reduction (L-SHADE) algorithm. Specifically, a mutual-information-based approach and the false nearest neighbours method are used to calculate the time delay and embedding dimensions, respectively. Then, the phase space of the wind speed time series is reconstructed based on the time delay and embedding dimensions, and the characteristics are analysed. The maximum Lyapunov exponent is applied to confirm the chaotic properties of the wind speed time series. Finally, EDNM trained by L-SHADE is used to predict the wind speed for Sotavento, which is located near Galicia, Spain. This study is the first, to the best of our knowledge, to use a dendritic neuron model to implement such real-world prediction. Extensive experimental results show that the proposed EDNM can perform better than other state-of-the-art models in terms of different assessment criteria. Therefore, the proposed method has high potential for practical applications in electric power systems.