Surrogate Model-Based Approximate Optimum Design with Discrete and Continuous Design Variables for the Package Installation Substructure of a 10 MW Offshore Wind Turbine

As offshore wind turbines continue to grow in capacity, reducing the structural weight of substructures has become a key factor in securing project economics, and the package installation method has attracted attention as a promising technology for reducing installation costs. This study presents an integrated design procedure for the package installation substructure of a 10 MW fixed offshore wind turbine, ranging from the establishment of classification rule-based design load conditions to surrogate model-based approximate optimization and verification of the optimum designs. Design load conditions for the installation, operation, and survival phases were defined in accordance with DNV classification rules and IEC 61400-3-1, and the structural safety of the initial design was evaluated via finite element analysis. Response data for 243 design matrices were generated using an orthogonal array design in which the thicknesses of nine primary structural members were defined as three-level design variables. Kriging, response surface methodology (RSM), and radial basis function neural network (RBFN) surrogate models were compared using multiple statistical metrics, and the RBFN, exhibiting the highest average coefficient of determination of 0.942 together with the lowest error levels for the stress responses, was selected for the approximate optimization. Discrete and continuous design variable optimizations were carried out in parallel by coupling the RBFN surrogate model with the adaptive simulated annealing (ASA) algorithm. The discrete optimum design reduced the structural weight by 2.5% while satisfying the allowable stress criteria for all load conditions, converged with approximately 94% fewer design evaluations than the continuous approach, and is defined directly in manufacturable plate thicknesses; it was therefore adopted as the final design.