Time-varying interval prediction and decision-making for short-term wind power using convolutional gated recurrent unit and multi-objective elephant clan optimization

Wind power (WP) interval prediction has attracted more and more attention in recent years due to WP's intermittency and uncertainty. However, traditional interval prediction models suffer from data distribution assumptions, fixed interval widths, etc. This paper proposes a hybrid WP interval prediction method to eliminate these constraints. First, the mean impact value is applied to select the optimal inputs from meteorological factors, which are considered together with the WP data. Then, we propose a time-varying interval optimization strategy to construct prediction intervals (PIs) and avoid the restrictive condition of data distribution, which can provide a time-varying interval width simultaneously. Meanwhile a convolutional gated recurrent unit is performed to extract the spatial-temporal features and generates the prediction results. Based on these results, the final PIs are generated by the proposed multi-objective elephant clan optimization, in which three objectives are optimized simultaneously. To evaluate the performance of the proposed model, two WP datasets in Ningxia, China, are used. Finally, the obtained PIs are applied for decision-making to offer planned productions in the future and estimate the operational costs. The comparison results indicate that the proposed model can provide high-quality PIs and achieve lower operational costs than the benchmark models in decision-making.