A New Multi Deep Learning Technique With MR-IG Input Selection Algorithm for Multi-Step Wind Forecasting

Abstract

In recent years, the use of renewable energy sources has increased significantly. Among these, wind energy stands out as abundant, cost-effective, highly efficient in energy conversion, and environmentally sustainable. This study proposes a hybrid approach based on advanced deep learning techniques for multi-step wind forecasting. The hybrid model integrates enhanced deep learning methods, optimal feature selection techniques, and decomposition transformation models to achieve precise multi-step wind forecasts. Our proposed method incorporates a sequence of robust techniques, including variational mode decomposition for signal discretization, maximum relevance interaction gain for selecting valuable input features, and a predictive model combining convolutional neural networks, gated recurrent units, and bidirectional long short-term memory. This integration leverages the strengths of each model while minimizing their limitations, resulting in improved efficiency and accuracy in forecasting. To evaluate the proposed method, wind power generation data from the Pennsylvania–New Jersey–Maryland (PJM) electricity market and wind speed data from the Favignana Island microgrid were analysed. The results of multi-step wind power forecasting demonstrate the hybrid model's commendable accuracy. For instance, in the PJM electricity market, the average mean absolute percentage error for 2018 ranges from 3.8401% for 1-h-ahead forecasting to 13.8123% for 12-h-ahead forecasting.