Multi-step Ahead Wind Forecasting Using Nonlinear Autoregressive Neural Networks

Multi-step ahead wind forecasting is a key consideration for wind farm owners operating in a competitive electricity market for assessing the reliability of a power plant and devising an optimal dispatch strategy to maximize their revenues. In this scenario, the accuracy of the forecasts and swiftness of the prediction process are the major factors. This paper presents an accurate and fast mechanism for wind forecasting up to six steps in future. Two different multi-step prediction strategies, namely, direct strategy and recursive strategy are used for this purpose. A nonlinear autoregressive neural network is developed to implement these techniques for wind speed time series. The developed method is evaluated in terms of standard performance indices via a thorough case study considering real-world wind speed data. The simulation results depict the efficacy of the proposed methodology as compared to a benchmark prediction model especially for longer time horizons.