A Reinforcement Learning Method for Intermediate Point Enthalpy Control in Super-critical Power Unit

摘要

The intermediate point enthalpy is a significant indicator for the steam temperature in thermal power unit, which has a great impact on the safety and economy in power unit. The intermediate point enthalpy control becomes more difficult in supercritical power unit than in subcritical unit because of controlled plant's characteristics of non-linearity, large inertia and coupling between different input. In this paper, a reinforcement learning method using proximal policy optimization algorithm is proposed to operate the feed water following control scheme for coordinated control system in supercritical unit. Experiment indicates that the proposed method can achieve satisfactory control performance compared with feed-forward decoupling control.