Multivariate interval wind power forecasting based on QRTCN-GRU optimized by IHOA

The randomness of wind power pose significant challenges for traditional point forecasting methods to accurately describe its uncertainty. Therefore, a multivariate interval wind power forecasting method based on quantile regression temporal convolutional network-gated recurrent unit (QRTCN-GRU) optimized by improved hiking optimization algorithm (IHOA) is proposed. Firstly, Pearson correlation coefficient is employed for feature selection. Then, a wind power forecasting method is constructed based on the strengths of temporal convolutional network in capturing local features and multi-scale dependencies in time series, combined with the long-term memory capability of gated recurrent unit, and the attention mechanism is applied to focuses on the influence of key information. Subsequently, quantile regression is employed to model the uncertainty of wind power, and the interval forecasting at different confidence levels is realized. Finally, the hyperparameters of QRTCN-GRU are optimized by IHOA to further improve its performance. Experimental results demonstrate that QRTCN-GRU exhibits superior performance in both forecasting precision and describing wind power uncertainty compared to other interval forecasting methods.