Usage of correlation analysis and hypothesis test in optimizing the gated recurrent unit network for wind speed forecasting

Accurate wind speed forecasting is vital to unlocking the potential of wind energy and improving its utilization. By focusing on the input optimization of the network, this study develops a novel wind speed forecasting system based on a deep learning gated recurrent unit (GRU) network. Specifically, to begin with, the Pearson correlation, the partial correlation, and the maximum information coefficient analyses were employed to extract the necessary input variables according to the correlation coefficients with large values. Later, a t-test was used to further select the variables that had small correlation coefficients with the concerned wind speed variable. Subsequently, auto- and partial auto-correlation analyses were adopted to determine the related hyperparameters of the network. Finally, the wind speed was forecasted by incorporating the selected input variables and hyperparameters into the GRU network. To assess the performance of the proposed system, apart from the usual error criteria, we adopted two additional hypothesis test approaches, that is, the Friedman and Nemenyi tests, to qualify the forecasting performance of networks with different input settings. The forecasting performance sorting results, as well as error comparison results at different stations against three other well-known models, demonstrate that the proposed wind speed forecasting system performs well.