Decoupling control of core power and axial power distribution for large pressurized water reactors based on reinforcement learning

The automatic control of core power and axial power distribution in large pressurized water reactors (PWRs) is crucial for reactor safety, which can be realized by the Mechanical Shim (MSHIM) control strategy. However, this strategy suffers from strong coupling between the regulations of reactor power and axial offset (AO), and the decoupling control between them is expected. This paper proposes a decoupling MSHIM control strategy for large PWRs based on reinforcement learning (RL). Two feedforward RL-agents are designed to compensate for the speeds of two independent control rod banks determined by the MSHIM control system. Thus, coupling effects between them can be eliminated when regulating the reactor power and AO. Simulation results of the AP1000 reactor under typical operational transients show that the RL-based decoupling MSHIM control strategy can provide tighter AO control than both the original and a conventional feedforward decoupling MSHIM control strategies with little or no compromise in reactor power control. The mean absolute percentage error and maximum absolute error of AO can be reduced by up to 99.7 % and 94.9 %, respectively, compared with the original strategy. This study provides an effective solution for decoupling core power and AO control in large PWRs under MSHIM control strategy.