A reinforcement learning based hybrid evolutionary algorithm for ship stability design

Over the past decades, various search and optimisation methods have been used for ship design – a dynamic and complicated process. While several advantages of using these methods have been demonstrated, one of the main limiting factors of optimisation applications in ship design is the high runtime requirement of the involved simulations. This severely restricts the number of real applications in this area. This chapter presents a hybrid evolutionary algorithm that uses reinforcement learning to guide the search. Through giving and correcting the search direction, the runtime of optimisation can be effectively reduced. The NSGA-II, a well known multi-objective evolutionary algorithm, is utilised together with reinforcement learning to form the hybrid approach. As an important optimisation application field, the ship stability design problem has been selected for evaluating the performance of this new method. A Ropax (roll on/roll off passenger ship) damage stability problem is selected as a case study to demonstrate the effectiveness of the proposed approach.