A dynamic parallelized hybrid optimization method for 22 MW large-scale floating offshore wind turbines

This study proposes a dynamic parallelized hybrid optimization method to achieve high-precision, AI-accelerated multi-objective optimization for 22 MW floating offshore wind turbines. The proposed method addresses the computational bottlenecks arising from time-domain coupled simulation during the evaluation stage, resolves slow optimization convergence issues, and enhances the exploration capacity of the solution space. Specifically, a dynamic parallelized task scheduling strategy is developed, which treats design samples within a population as batch tasks. By dynamically allocating independent processes, this strategy enables the automatic parallel evaluation of multiple design samples. Compared with traditional optimization methods, this strategy achieves 3.87 times efficiency enhancement in population evaluation. Furthermore, a hybrid optimization module integrating incremental learning neural networks with optimization algorithms is developed to achieve the automatic enhancement of predictive accuracy and co-optimization with the core iteration process. This module facilitates the optimization process by pre-optimizing new populations and converting defective design samples into predicted optimal ones, thereby achieving 7 times increase in the quantity of optimized samples. Integrating the dynamic parallelized task scheduling strategy with the hybrid optimization module improves overall convergence efficiency by a factor of 24.51. Consequently, this method provides robust technical support for the engineering design optimization of large-scale floating offshore wind turbines.