A NOVEL REINFORCEMENT LEARNING FRAMEWORK FOR PROPULSION SYSTEM OPTIMIZATION

— Enhancing propulsion system efficacy presents a significant challenge in dynamic settings, where the pursuit of high thrust frequently contradicts the objective of reducing fuel consumption. This study presents an innovative reinforcement learning (RL) framework utilizing the Deep Deterministic Policy Gradient (DDPG) algorithm to effectively manage this trade-off. Through the creation of a bespoke simulation environment, the propulsion system is represented with state and action spaces that precisely reflect thrust and fuel dynamics, allowing the reinforcement learning agent to acquire effective, continuous control policies suited to the intricacies of the problem. Significant contributions are offered by this research in the form of a customized propulsion system environment, a thorough evaluation under nominal conditions, and the implementation of a DDPG framework for continuous control. The DDPGbased approach achieves a better balance between thrust optimization and fuel efficiency than traditional methods, as shown in a comparison study. Aerospace, marine, and industrial propulsion systems stand to benefit greatly from these results, which highlight the promise of reinforcement learning in solving complex engineering problems.