Evolutionary Reinforcement Learning-based Optimization Method for Ship Concept Schemes

This paper proposes an efficient optimization algorithm for ship conceptual design solutions. The algorithm is based on Evolutionary Reinforcement Learning (ERL) and is designed to address the issues of slow optimization speed and unclear optimization results in traditional optimization algorithms. It utilizes expert preferences to design intelligent body environment, state, action, and reward models based on the parameter library of the ship concept design. An improved ERL algorithm is developed to enhance optimization speed and optimization stability, and simulation results demonstrate that the proposed improved ERL model architecture and experts' experience can optimize multiple ship conceptual design solutions within 7.5 seconds. The optimization results obtained from the subjective and objective evaluation show 14.61% and 11.11% improvements over the traditional optimization method, respectively. Compared to the traditional optimization methods NSGA-II and ERL algorithm, the improved algorithm achieves a 59.57% and 5.63% improvement, respectively. The proposed algorithm can serve as a valuable reference for optimizing ship conceptual design schemes.