Comparison of Three Methods for Short-Term Wind Power Forecasting

Abstract

Wind power forecasting is critical for effective grid operation and management. An accurate short-term wind forecasting model is an important tool for grid reliability and market-based ancillary services. However accurate prediction of wind power is not a trivial task. This is mainly because wind is stochastic in nature and a very local phenomenon, and therefore hard to predict. In this paper, we compared three methods for short-term wind power forecasting. Namely, a time series based method called Autoregressive Moving Average (ARMA), Artificial Neural Networks (ANNs), and a method based on hybridising Artificial Neural Networks (ANNs) and Fuzzy Logic called Adaptive Neuro-Fuzzy Inference Systems (ANFIS). It is shown that for a very short-term wind power forecasting, all the three methods perform similarly. However, for the short-term wind power forecasting, the ARIMA method performs better than both the ANNs and ANFIS. For longer time horizon (medium and long-term), the performance of ARMA deteriorated as compared to the other two methods.