Designing an Optimal PID Controller for a Gas Turbine System Using Reinforcement Learning

This paper investigates the application of reinforcement learning (RL) techniques for optimizing proportional–integral–derivative (PID) controller parameters in gas turbine speed control systems. The research employs the Rowen mathematical model as the foundational framework and introduces a novel approach utilizing twin‐delayed deep deterministic policy gradient (TD3) algorithms. The methodology integrates machine learning with classical control theory to address the persistent challenges of maintaining optimal turbine speed during both transient startup phases and steady‐state operations. Implementation was conducted using a simulation environment based on MATLAB/Simulink, with the General Electric 5001M heavy‐duty gas turbine serving as the reference system. The RL agent was designed to interact with the simulated environment, continuously refining controller parameters to minimize performance metrics including integral error values, rise time, and settling characteristics. Comparative analysis between the proposed TD3‐optimized PID controller and conventional tuning methods demonstrates significant performance enhancements across multiple control criteria. The optimized system achieved notable reductions in settling time, overshoot magnitude, and steady‐state error, while also demonstrating improved disturbance rejection capabilities under variable load conditions and sensor noise.