PI Feedforward Control with Reinforcement Learning Approach for Coupled Tank System

In this study, a hybrid approach, PI (ProportionalIntegral) Feedforward control with Reinforcement Learning, is designed to control the liquid level on a single-input single-output coupled tank system. For linear systems, PI feedforward control offers simple design and effective response, but in the presence of nonlinearities, disturbances, and time delays, its performance degrades. To address this issue, an RL approach in feedback is integrated with the PI feedforward control to adjust control actions of the coupled tank system. To enhance system response and robustness, the control signals from the RL agent and PI-Feedforward are integrated. The simulation results show that the combined control strategy offers improved transient response and steady-state accuracy compared to PI feedforward control and fuzzy PI feedforward control techniques, especially under parameter variation and disturbances.