Research on intelligent hierarchical control method for U-tube steam generator water level based on TD3-MPC

Abstract

Aiming at the issues of poor control performance at low power levels and difficulties in precise modeling for U-tube steam generators (UTSG), this paper proposes an intelligent hierarchical control method optimized based on deep reinforcement learning. The primary controller adopts the model predictive control (MPC) algorithm, which directly regulates the water level through an improved objective function. The advanced controller is an agent-based controller utilizing the model-free twin delayed deep deterministic policy gradient (TD3) algorithm, responsible for real-time optimization of the MPC controller parameters. Additionally, to enhance the learning efficiency and convergence performance of the agent, a reward function design framework based on nonlinear scaling and normalization is proposed. Simulation results demonstrate that, compared to traditional methods, the proposed approach achieves precise control of UTSG water level, significantly improves control performance, and enhances adaptive capability and system robustness.