Disturbance Rejection in Quadruple-Tank System by Proposing New Method in Reinforcement Learning

This paper aims to propose a new method for reinforcement learning and compare it with a PID controller in the Quadruple-tank system in the presence of uncertainty. We use one of the popular structures called actor-critic and train it using a deep deterministic policy gradient algorithm. These methods are compared in terms of accuracy and rise time to show which one can have better performance if we consider uncertainty. The proposed method represents an approach that considers some changes in observation dimensions by a series of error items consisting of some previous and current errors and then trains the Reinforcement Learning algorithm through these new observations. Finally, the results of these methods are compared by simulations and the proposed method's performance is evaluated. Which specifies our approach has better performance.