Smart Operation Guidance for Ship Engine Rooms: Integrating Bayesian Networks and Reinforcement Learning

Ship engine rooms are critical propulsion centers requiring precise operational coordination for maritime safety and efficiency. Current manual operations rely heavily on crew expertise, creating challenges in achieving consistent accuracy and real-time adaptability in complex marine environments. This study integrates expert knowledge-based Bayesian networks with PPO+LSTM reinforcement learning to develop intelligent operation guidance systems. A 42-node Bayesian network model captures inter-equipment operational logic using domain expertise and marine simulators, while the PPO+LSTM agent learns optimal strategies through environment interaction, leveraging LSTM's sequential memory for complex temporal dependencies. Experimental results demonstrate superior performance: 91.2% success rate, 83.5 average score, and 47 -step completion, significantly outperforming DQN (77.9% success rate) and standard PPO (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$88.7\%$</tex> success rate). This research advances maritime automation by providing a systematic framework for intelligent ship operations, potentially reducing human error and enhancing operational safety.