Application of deep learning in wind, solar, and ocean energy: An analysis of prediction, optimization, and operation & maintenance

This study conducts a systematic review and coding analysis of deep learning (DL) applications in renewable energy systems (wind, solar, and ocean) from 2012 to 2024, addressing global sustainable development needs. We develop a comparative framework structured as “energy type (wind/solar/ocean) × task type (prediction/optimization/operations and maintenance),” with model families (LSTM, CNN, Transformer) as the third dimension, to systematically assess differences and similarities in data requirements, generalization ability, and computational overhead. Key findings indicate explosive growth of DL in solar and wind energy, with mature applications in short-term prediction and operational optimization; ocean energy advances slowly, with marked deficiencies in operations and maintenance (O&M) research. Model suitability shows that LSTM and CNN exhibit robustness in short-term time-series forecasting and fault detection; Transformers perform well in high-dimensional, multivariate, long-sequence scenarios but falter under data scarcity or domain shifts. The contributions of this work are: (1) a unified cross-energy, cross-task, cross-model comparative framework; (2) identification of systemic O&M gaps in ocean energy; (3) extraction of reusable model–task matching principles and constraints; and (4) a proposed ocean energy research roadmap emphasizing multisource sensor fusion, transfer/self-supervised learning, and physics-informed data-driven integration to improve real-time capability and intelligence. This framework offers structured evidence and methodological guidance for integrated modeling, scheduling, and O&M in multi-energy systems.