An AI-Based Hybrid Forecasting Model for Wind Speed Forecasting

Abstract

Forecasting of wind speed plays an important role in wind power prediction for management of wind energy. Due to intermittent nature of wind, accurately forecasting of wind speed has been a long standing research challenge. Artificial neural networks (ANNs) is one of promising approaches to predict wind speed. However, since the results of ANN-based models are strongly dependent on the initial weights and thresholds values which are usually randomly generated, the stability of forecasting results is not always satisfactory. This paper presents a new hybrid model for short term forecasting of wind speed with high accuracy and strong stability by optimizing the parameters in a generalized regression neural network (GRNN) using a multi-objective firefly algorithm (MOFA). To evaluate the effectiveness of this hybrid algorithm, we apply it for short-term forecasting of wind speed from four wind power stations in Penglai, China, along with four typical ANN-based models, which are back propagation neural network (BPNN), radical basis function neural network (RBFNN), wavelet neural network (WNN) and GRNN. The comparison results clearly show that this hybrid model can significantly reduce the impact of randomness of initialization on the forecasting results and achieve good accuracy and stability.