Deep reinforcement learning-based approach for control of Two Input–Two Output process control system

This study investigates the use of a Deep Deterministic Policy Gradient (DDPG) algorithm to control a multivariable coupled system, specifically a two input–two output (TITO) system. Traditional control methods, such as proportional–integral–derivative (PID) controllers and decoupling techniques, often face limitations in handling the complex, nonlinear dynamics and interactions within Multi Input Multi Output (MIMO) systems. The DDPG-based approach, leveraging the actor-critic architecture for continuous action spaces, enables adaptive policy learning and robust performance. Experimental results demonstrate that the DDPG controller performs significantly well compared with conventional controllers, achieving minimum integral squared error (ISE), integral absolute error (IAE), and integral time of absolute error (ITAE), indicating superior performance in minimizing deviations from target levels. These findings highlight the potential of deep reinforcement learning (DRL) for advanced multivariable control, suggesting avenues for future applications in larger and more intricate industrial systems.