A novel ultra-short-term wind power prediction model jointly driven by multiple algorithm optimization and adaptive selection

Abstract

Ultrashort-term wind power forecasting with great precision and robustness is essential for improving power quality and reliability management and reducing the cost of rotating backup supply, thus guaranteeing the security and stability of power systems in large-scale grid-connected wind power. This study proposed a novel ultra-short-term wind power prediction model jointly driven by multiple algorithm optimization and adaptive selection. The original wind power sequence is decomposed into smooth subsequences by the optimized variational mode decomposition algorithm. Each sequence is predicted in advance by two outstanding prediction methods. The method with high accuracy is automatically selected for the prediction output of that sequence. The two excellent models are least square support vector machine optimized by improved whale optimization algorithm and hybrid kernel extreme learning machine optimized by sine cosine search-sparrow search algorithm, improving the prediction accuracy and efficiency. Based on three publicly available datasets, the proposed model has more than 41 % percent improvement in root mean square error compared to the current studies and about 20 % percent improvement in root mean square error compared to the proposed models without selection strategy. Combined with the adaptive selection concept, the proposed prediction model can obtain more accurate wind power prediction results with higher prediction accuracy, more substantial prediction generalization, and robustness.