Deep-Learning-Based Scheduling Optimization of Wind-Hydrogen-Energy Storage System on Energy Islands

With the growing global demand for climate change mitigation, the development and utilization of renewable energy have become crucial for energy transition. This study introduces an innovative optimization framework for clean energy systems on energy islands, integrating offshore wind power, hydrogen production, and hydrogen storage. Advanced forecasting models based on Long Short-Term Memory (LSTM) and Attention-enhanced Convolutional Neural Networks combined with Bidirectional LSTM (Attention-CNN-BiLSTM) are proposed, achieving an impressive prediction accuracy of 98 % for both wind power and residential electricity load. A multi-objective optimization approach, combining the Non-dominated Sorting Genetic Algorithm II (NSGA-II) with the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS), is employed to perform 24-h rolling scheduling optimization of the energy system. The optimization model finds a compromise between maximizing profits and minimizing power fluctuations. Compared with the results of non-optimization, the power stability of the optimized system is improved by 45 %. When the wind power capacity is sufficient, the system operating profit reaches 4.41 million CNY, and the power fluctuation is 4.26 GW. This study provides a new theoretical basis and practical guidelines for the design and operation of energy islands, highlighting the potential applications of clean energy technologies in modern energy systems.