Adaptive tuning of fractional order PID controllers for nonlinear processes using hybrid PSO DQN reinforcement learning

This study presents an innovative adaptive non-linear fractional-order PID (FOPID) tuning methodology for a flow meter controller in a desalination plant, integrating a hybrid Particle Swarm Optimization (PSO) and Deep Q-Network (DQN)-based Reinforcement Learning (RL) strategy with a dynamic weighting mechanism to optimize control of non-linear systems with time delays and disturbances. By utilizing fractional-order parameters, the PSO-DQN-RL framework ensures global optimization and real-time adaptability under fluctuations in operational parameters. Results demonstrate superior performance over traditional methods and advanced techniques such as Genetic Algorithms (GA), Fuzzy Logic Controller (FLC), Neural Network-based PID (NN-PID), and PSO, offering faster response times, reduced overshoot, and minimal steady-state error compared to the slower and less precise outcomes of FLC, the static limitations of PSO, the rigid parameter settings of GA, and the inconsistent performance of NN. The hybrid method's enhanced robustness and dynamic parameter evolution surpass the modest adaptability of PSO. Despite its computational complexity, the offline-online balance and real-time GUI enable scalable deployment, positioning this scientifically novel approach as a benchmark for FOPID tuning in various applications.