Wind power forecasting by a new local quadratic wavelet neural network

This paper proposes a new local quadratic wavelet neural network (LQWNN) for short-term wind power forecasting in electrical power systems. Due to variability of wind power, its integration into the power grid requires an exact and accurate prediction of wind power. The proposed approach in this paper can effectively model the nonlinear behavior of the wind power, owing to nonlinear structure of its local quadratic models (LQM). Localized in both time and frequency domains, the wavelet functions in the LQWNN set the validity regions of the LQMs, resulting in learning efficiency and structure transparency of the overall model. The proposed LQWNN approach is training using a simple, but efficient particle swarm optimization (PSO) algorithm and its parameters (i.e. parameters of LQMs and wavelet functions) are optimally tuned. Performance of the proposed approach for wind power forecasting is assessed through two real-world case studies, namely wind power forecasting in Irish grid, Ireland and Waubra wind farm, Australia. Analysis of the obtained results and comparison to other methods indicates the promising performance of the proposed LQWNN for wind power forecasting applications.