Wind Power Prediction Based on Nonlinear AutoRegressive with eXogenous Inputs Algorithm

In modern power systems with an increasing proportion of renewable energy sources, long-term wind speed forecasts are essential for grid scheduling due to the volatility of wind speeds. In this paper, we use stepwise regression to determine the influencing factors of wind speed and their weights, and then fit the influencing factors through a Nonlinear AutoRegressive with eXogenous inputs (NAR) neural network to accurately predict the wind speed. We improved the traditional NAR method by dividing it into two modes: open-loop and closed-loop, and added the closed-loop feedback to make the model predictions more accurate. The experimental results show that compared with the traditional recurrent neural network (RNN) and complex-valued feedforward neural network (CFNN), the proposed prediction model has a higher mean-square error of 1.045644e-5 compared with the RNN, and a lower mean-square error of 2.505524e-4 compared with the CFNN. In terms of computational cost, this scheme is significantly better than the other two schemes, requiring only 5.4% and 14.2% of the computation and training time of the other two methods, respectively.