The application of heat-pipe reactor (HPR) coupled with supercritical CO₂ (sCO₂) Brayton cycle has been identified to have attractive potential due to its advantageous features, such as high cycle efficiency, passive safety and compactness. However, there remains limited investigation on the application of intelligent control methods to optimize coordinated control actions of the reactor and the energy conversion system during load changes and system disturbances while considering effective load-following and safe thermal transients. This study, therefore, applies Reinforcement Learning (RL) through safety-constrained twin-delayed deep deterministic policy gradient (TD3) controller acting through six available control inputs: shaft-speed command, precooler cooling-water flow, compressor make-up flow, compressor bleed flow, hot-side bypass flow and reactor reactivity to optimise the control of a RELAP5-based HPR–SCO₂ system while restricting actions that could lead to unacceptable thermal or operating conditions. Performance is evaluated under load changes and representative reactor and cycle disturbances and then compared with the original coordinated control system in terms of load-following, reactor–heat-removal coordination, mitigation of temperature excursions, thermal gradients and thermal ramp rates, with the ultimate goal of enhancing the safety and robustness of intelligent control for operation of HPR systems.
 

Heat-pipe reactor; sCO2 Brayton cycle; safety-constrained reinforcement learning; abnormal transients

暂无