Model Predictive Control for Gas Turbine based on Proximal Policy Optimization

Gas turbines are essential for energy conversion, and their efficient and economical operation is vital for industrial energy savings and emission reductions. To address the challenges of reduced control accuracy and decreased stability in heavy-duty gas turbines during changes in operating modes, we introduce a self-tuning strategy for model predictive control parameters using the proximal policy optimization approach, which incorporates a shear term to constrain the amplitude of policy updates during the policy gradient update process, significantly enhancing training stability and sampling efficiency compared to traditional reinforcement learning approaches, which leads to a more stable and efficient policy learning process. To validate the effectiveness of the proposed approach, performance tests were conducted on gas turbines under various load modes and compared with other model predictive control parameter tuning approaches. The test results demonstrate that model predictive control parameterized with this strategy achieves improved accuracy and response speed, ensuring stable operation of the gas turbine across different control modes.