Prediction of mooring tension of floating offshore wind turbines by CNN-LSTM-ATT and Chebyshev polynomials

Offshore wind technology has emerged as a promising solution to exploit wind resources in deep waters. For floating offshore wind turbines, mooring systems are critical in maintaining station-keeping functions. Effective monitoring of mooring loads is crucial for ensuring safe and cost-efficient operation and maintenance. However, direct measurement of mooring line tensions is often costly. To address this challenge, this paper focuses on predicting mooring tensions using accessible motion data from the platform. We propose employing deep learning algorithms, specifically the advanced CNN-LSTM-ATT neural network model, to capture six-degree-of-freedom platform motions from Orcaflex under extreme sea conditions. This model shows robust predictive performance in most sea conditions. However, it faces significant challenges when dealing with short wave periods, where the strong nonlinearity and non-stationarity of the system hinder the neural network’s ability to accurately capture features and predict mooring line tensions. To overcome these challenges, we introduce two optimization methods that incorporate Chebyshev polynomials, renowned for their quick and effective capture of overall trends, particularly fitting low frequencies. These methods significantly enhance the performance of the CNN-LSTM-ATT model. The findings contribute to the real-time prediction and structural health monitoring of floating wind turbines, thereby enhancing their safety and reliability.