Ultra-Short-Term Offshore Wind Power Prediction Based on Informer with Hybrid Data Preprocessing

Abstract

The wind power prediction contributes significantly to the reliable operation of wind energy generation systems and the efficient scheduling of integrated power grids. However, the intrinsic characteristics of wind energy make the precise prediction of wind power remain a complex issue, and existing approaches still show some disadvantages when facing with large-scale data. To cope with this problem, a novel deep learning (DL) algorithm called Informer is applied to the ultra-short-term offshore wind power prediction in this research, which can extract features and capture the sequence dependency more effectively from long time-series data. In addition, the discrete wavelet transform (DWT) and improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) are adopted to the data preprocessing to achieve more precise prediction. Prediction results of the Informer-based approach are analyzed and contrasted with benchmarks including the long short-term memory (LSTM)-based, DLinear-based, and Transformer-based approaches under varying lengths of input sequence prediction and multi-time steps ahead prediction. Results show that the proposed approach is superior to the benchmarks for all diverse instances studied in this research.

Keywords: Offshore wind power prediction, Ultra-short-term, deep learning, Informer, Data preprocessing