A method for quickly setting parameters of model predictive control of a three-shaft gas turbine based on transfer learning
By constructing a linear variable parameter prediction model and a model predictive controller based on transfer learning, and combining the soft actor commentator algorithm and transfer learning mechanism, adaptive optimization and rapid tuning of model predictive control parameters for a three-axis gas turbine are achieved. This solves the problems of low parameter tuning efficiency and poor repeatability in existing technologies, and improves control performance under multiple operating conditions.
Patent Information
- Application Number
- CN202610676193.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-16
- Publication Date
- 2026-07-17
AI Technical Summary
Existing model predictive control methods for three-axis gas turbines struggle to simultaneously achieve dynamic response speed, steady-state tracking accuracy, overshoot suppression capability, and control smoothness under various operating conditions, strong constraints, and multi-source uncertainties. Furthermore, they suffer from low parameter tuning efficiency, poor repeatability, and reliance on manual experience and repeated debugging.
A linear variable parameter prediction model and a model predictive controller are constructed using a transfer learning-based approach. By combining the soft actor critic algorithm and the transfer learning mechanism, adaptive optimization and rapid tuning of the model predictive control parameters are achieved. The weight configuration is adjusted online through reinforcement learning, and the parameters are reused under different operating conditions.
It improves the overall control performance of the three-shaft gas turbine under complex operating conditions, reduces the burden of manual tuning, enhances the efficiency of parameter configuration and adaptability across operating conditions, and improves the control quality of the controller under multiple constraints.
Smart Images
Figure CN122411030A_ABST