Prediction and multi-objective optimization of a floating wind-wave hybrid system using surrogate modeling
Abstract
Artificial intelligence techniques are revolutionizing the analysis of marine renewable energy systems. In this study, a surrogate model was employed to predict the dynamic response and power performance of a wind-wave hybrid system under operational conditions. The surrogate model was selected based on a comparison of the predictive performance of three machine learning models: Gaussian Process Regression, Multilayer Perceptron artificial neural networks, and Support Vector Regression, all of which were trained on a numerical simulation database. Test results demonstrate that the MLP-based surrogate model provides high-accuracy predictions, with the coefficient of determination exceeding 0.96 for all ten output variables considered. The model predictions indicate that a floating offshore wind turbine platform could be stabilized by appropriately configuring the power take-off of attached wave energy converters. Compared with conventional numerical simulation, the surrogate model substantially lowers the computational overhead. Facilitated by the high efficiency of the model, a genetic algorithm was applied to optimize both the dynamic response of the platform and the power performance of the wave energy converter array through adjusting power take-off settings. This paper offers an efficient methodology for the performance evaluation and optimization of floating wind-wave hybrid systems.