Optimization of a mooring system applying a deep neural network under multi-directional environmental conditions

Abstract

This study investigates the optimization of a mooring system using a deep learning technique, addressing the limitation of considering only a single environmental direction. To find a robust optimal design, a Deep Neural Network (DNN) model was established to predict mooring line tension and offset of the floating structure based on variations in component parameters and multiple environmental directions. The trained DNN model was then integrated with a Non-dominated Sorting Genetic Algorithm II (NSGA-II). A representative optimal solution focused on tension reduction showed a decrease of approximately 2.7 % compared to a base case, while another representative solution focused on offset reduction achieved a decrease of approximately 17 %. Furthermore, the most economical solution reduced the mooring line weight by approximately 19 %. A comparison confirmed that a mooring system designed from a single-direction optimization violated design constraints when its responses were evaluated under other environmental directions. This study confirms the feasibility of applying a deep learning technique to the mooring system optimization process and highlights the necessity of considering multi-directional environmental conditions to find a robust optimal design, while also significantly improving computational efficiency by approximately 50 %. Future work includes analyzing non-collinear environmental conditions and applying the methodology to various mooring configurations.