Reinforcement-Learning-Based Level Controller for Separator Drum Unit in Refinery System

The Basrah Refinery, Iraq, similarly to other refineries, is subject several industrial constraints. Therefore, the main challenge optimize parameters of level controller process unit tanks. In this paper, a PI designed for these important processes in which separator drum (D5204). Furthermore, improvement achieved under constraints, such as inlet liquid flow rate tank (m2) and valve opening yi%, by using two different techniques: first one conducted closed-Loop PID auto-tuner that based on frequency system estimator, via reinforcement learning approach (RL). RL employed through approaches: calculating optimal an offline tuner, second online tuner parameters. case, works PI-like RD5204. mathematical model RD5204 derived simulated MATLAB. Several experiments are validate proposed controller. Further, performance evaluated disturbances noise, results indict superior methods. increases controller’s robustness against uncertainty perturbations.