Deep Reinforcement Learning for Optimizing Electronically Controlled Propulsion: A DDPG-Based Approach

ABSTRACT Optimizing propulsion systems in dynamic environments is a difficult task that necessitates a delicate balance between increasing thrust and reducing fuel consumption. This paper presents an innovative reinforcement learning framework utilizing the Deep Deterministic Policy Gradient (DDPG) algorithm to tackle this trade-off. A bespoke simulation environment was created to emulate authentic propulsion system dynamics, integrating continuous state and action spaces that represent thrust, fuel efficiency, and environmental perturbations. The proposed method allows the reinforcement learning agent to develop adaptive control policies that surpass conventional techniques, like PID controllers, in attaining enhanced fuel efficiency and thrust stability. The DDPG framework exhibits substantial enhancements through thorough assessment in various operational contexts, attaining a cumulative reward increase of up to 40% and an 18% enhancement in fuel efficiency relative to traditional control methods. This study emphasizes the transformative capacity of reinforcement learning in enhancing propulsion system efficacy for aerospace, marine, and industrial applications, facilitating the development of sustainable and intelligent transportation technologies.