Water Content Control in Tobacco Leaf Loose and Moisture Regain Process based on Deep Reinforcement Learning

In this paper, the control problem of outlet moisture content is investigated for the tobacco leaf loosening and moisture regain process. Unlike existing strategies that heavily depend on manual expertise for adjusting water addition and hot air temperature, a novel control framework is proposed in this work. This framework integrates an optimal outlet moisture content control method and a traditional Proportion-Integral-Differential (PID) control loop, thereby reducing the dependence on manual intervention. Specifically, the hot air temperature and water addition flow rate are dynamically adjusted using the soft actor-critic (SAC) algorithm. Simulation results demonstrate that our proposed method can maintain the outlet moisture content with a steady-state precision of 0.1%.