A Cooperative Control Method Based on Reinforcement Learning

With the increase of system dimensions, the difficulty of decoupling increases exponentially and the controller becomes more difficult to design. A reinforcement learning framework combined with tradition control method is presented to get rid of decoupling trouble and simplified controller design. A Markov decision process of basis weight and moisture control is explicitly defined. A first-order Pade approximation is applied to solve the time delay problem in reinforcement learning. With the proposed design, the control scheme could obtain a good performance of the control system as same as the PID & feedforward decoupling control through cooperative control. Numerical simulations are given to demonstrate the effectiveness of the proposed method.