Reinforcement Learning Tuned PI Controller for Two Tank Interacting Hybrid System

This research work analyze the use of a Deep Neural Network based Reinforcement Learning (RL) method applied to a highly non-linear system to determine the optimized PI controller gains. The twin-delayed deep deterministic policy gradient (TD3) reinforcement learning algorithm is used to adjust the gains of a PI controller. The conventional PI tuning method requires a model of the system and knowledge about the dynamic behavior of closed-loop systems. Improper tuning of PI controller gains leads to in-efficient set point tracking, overshoot, and poor response time. In Recent trends, machine learning is used to find out the optimal controller gains for PI controllers. In this research, Twin Delayed Deep Deterministic Policy Gradient (RL-TD3 Agent), an actor-critic reinforcement learning methodology based on-policies is used to solve nonlinear problems, in which continuous spaces, as opposed to discrete ones, are where states and activities are located. Deep neural networks are used to carry out the agent's policy in the actor-critic training paradigm of deep reinforcement learning. The purpose of employing neural network topologies for reinforcement learning is to strengthen the link that results in an improved policy by obtaining reward signals. Multiple-Input Multiple-Output (MIMO) system known as the Two Tank Interacting Hybrid System(TTIHS) is used as a case study to implement the proposed method. The RL Agent (RL-TD3) is trained using real-time data obtained from the Real Time physical setup.