A Methodological Approach to Short-Term Coastal Wave Prediction Using Combined Pre-Analysis and Machine Learning

Abstract Wave energy conversion, transforming ocean wave energy into electricity, is a promising clean energy technology. Effective site selection requires assessing wave resource potential, safety, and economic feasibility, guided by accurate prediction models. For unmonitored sea areas, leveraging adjacent observation data bridges data gaps and enhances research applicability. While traditional models like SWAN rely on extensive datasets, machine learning methods, particularly Long Short-Term Memory (LSTM) networks, offer cost-efficient solutions. However, applications to unobserved regions and extended forecasts remain underexplored. Japan’s NOWPHAS network, with over 80 observation sites, provides extensive long-term wave data critical for predicting wave height and period. This study utilizes NOWPHAS data to analyze wave propagation and time delays, developing a framework for 24-hour predictions at observed and unobserved sites. The proposed methodology optimizes resource allocation and enhances the reliability of wave condition forecasts, supporting coastal development and the advancement of wave energy systems.