ANN–generated initial populations for PSO–based design of OWT jackets
Year: 2025
Type: article
Abstract: • Strategies for generating initial populations in autonomous jacket design for OWTs. • PSO as search algorithm, coupled with FEM and ANN as structural models. • Two starting populations are refined using ANNs through three different strategies. • ANN-refined and pre-computed solutions improve optimization quality and quantity. This paper investigates and proposes efficient strategies for generating initial populations in the automated design of jacket-type support structures for offshore wind turbines. The particle swarm optimization algorithm is employed as search and optimization method, while a finite-element-based model is used to evaluate the structural feasibility in the design process. This model computes the loads acting on the structure, assesses its structural response, and verifies key design requirements. Soil-structure interaction is also considered to account for foundation flexibility. A key contribution of this study is the use of an artificial-neural-network-based surrogate model to estimate the structural utilization factor during the initial population generation phase. Since high-fidelity evaluation is not essential at this early stage, the neural network is used for its ability to rapidly estimate the structural performance. The obtained candidates satisfy a wide range of criteria, including ultimate limit states, fundamental frequency checks, joint and geometric verifications, and foundation requirements. Several strategies are proposed for generating initial populations in a pre-optimization phase. Results demonstrate that these strategies significantly increase not only the number of feasible designs but also their quality, measured in terms of minimal material usage and compliance with design criteria. The overall algorithm performance is substantially improved.
Source: Ocean Engineering
Institution: Universidad de Las Palmas de Gran Canaria
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $4,150
OpenAlex ID: W4415545219