Wind Speed Multi-Step Forecasting Using High- Resolution and Multi-Objective Ensemble Model With Error Correction

Abstract

Accurate wind forecasting is essential for enhancing the stability and efficiency of wind power systems. However, the nonlinear and unstable characteristics of wind poses significant challenges to achieving high-performance predictions in this domain. To address the issues of insufficient wind speed data decomposition and the limited forecasting time range for high-resolution data in hybrid prediction models, this study proposes a short-term, multistep wind speed prediction model based on high-resolution data and multi-objective integration with error correction. Initially, the White Shark Optimizer (WSO) algorithm is employed to determine the optimal decomposition parameters for Variational Mode Decomposition (VMD), decomposing the original high-resolution wind speed data. Secondly, the decomposed data is processed using Convolutional Neural Networks (CNN) with three different sizes of convolutional kernels to capture multiscale features, which are then fed into a Gated Recurrent Unit (GRU) model for three-step forecasting. Finally, three-step prediction results for features across different scales are input into an ensemble module, and the Non-Dominated Sorting Whale Optimization Algorithm (NSWOA) is utilized to tune the weight parameters of the ensemble and error correction (EC) module. Experimental results on different datasets indicate that the proposed approach not only leverages the detailed information from high-resolution data but also addresses the issue of low accuracy in multistep forecasting. Moreover, the hybrid model, which includes a multiobjective optimization-based integration module and error correction, not only provides highly accurate multistep forecasts but also ensures greater model stability.