Towards Safe and Efficient Operation of Floating Offshore Wind Turbines through Machine Learning-based Wave Forecasting

Abstract

In recent years, the global shift toward carbon neutrality has accelerated the deployment of offshore wind power, particularly floating wind turbines in deep coastal waters such as those surrounding Japan. These structures require regular maintenance, which is often conducted several kilometres offshore, where weather and sea conditions are highly variable and difficult to predict. Accurate wave forecasting is essential to ensure safe and efficient maintenance operations, especially when planning activities three to seven days in advance. This study investigates the application of machine learning techniques to predict significant wave height, wave period, and wave direction using long-term observational data from Japan’s Nationwide Ocean Wave information network for Ports and HArbourS (NOWPHAS). A Long Short-Term Memory (LSTM) model was trained with datasets spanning one to nine years, and its prediction accuracy is evaluated across multiple time horizons. The results demonstrate a strong negative correlation between the volume of training data and prediction error, with nine year datasets yielding the most reliable forecasts. Furthermore, the study proposes a decision-support framework for offshore maintenance scheduling based on multi-phase wave parameter predictions. The findings suggest that machine learning-based predictive models can support safer, more efficient maintenance planning for floating offshore wind turbines.