Forecasting Wind Speed Time Series Via Dendritic Neural Regression

Abstract

Wind energy is considered one of the fastest growing renewable ('green') energy resources. Precise wind power forecasting is imperative to ensure reliable power system planning and wind farm operation. However, traditional methods cannot yield satisfactory forecasts because of the chaotic properties and high volatility of wind speed time series. To address this issue, the use of artificial neural networks has attracted increasing attention owing to their significantly enhanced prediction accuracy. Based on these considerations, a novel neural model with a dynamic dendrite structure, known as the dendritic neuron model (DNM), can be adopted for wind speed time series prediction. The DNM is a plausible biological neural model that was originally designed for classification problems; accordingly, this study proposes the use of a regressive version of the DNM, named dendritic neural regression (DNR), in which the dendrite strength of each branch is considered. To enhance the prediction performance, the recently proposed states of matter search (SMS) optimization algorithm is used to optimize the neural architecture for DNR. By virtue of the powerful search ability of the SMS algorithm, DNR can efficiently capture the nonlinear correlations among distinct features and dendritic branches. Extensive experimental results and statistical tests demonstrate that compared with other state-of-the-art prediction techniques, DNR can achieve highly competitive results in wind speed forecasting.