Physics-Informed Neural Network-based PID Control Design: Application to Benchmark Problems

In this study, first-, second-, and fourth-order benchmark control systems were implemented to determine the optimal proportional-integral-derivative (PID) gains Kp, Ki, and Kd by minimizing a physics-informed loss function using a physics-informed neural network (PINN), enabling stable convergence to the target value (Setpoint). The derived PID gains were validated in MATLAB/Simulink under various cases. Simulation results using PID gains obtained from conventional PID, PID-neural network (PID-NN), and PINN-PID methods demonstrated that the proposed PINN-PID approach significantly improved convergence stability and consistently stabilized the system at the target value, exhibiting superior control performance compared to the other techniques.