Application of RBF neural network PID control on buck DC-DC converter

Abstract A structure to solve low control accuracy and poor tracking performance in traditional PID controlled DC-DC converters when input voltages and load currents fluctuate has been proposed: combining PID control with Radial Basis Function (RBF) neural networks. To optimize the PID controller’s parameters, this approach makes use of the RBF neural network’s approximation capability for nonlinear functions. Under different test conditions, the neural network takes the output voltage and its deviation from the reference value as inputs. The hidden layer employs Gaussian activation and adjusts by gradient descent. The output layer computes the PID controller’s Proportional (Kp), Integral (Ki), and Derivative (Kd) by the linear combination of hidden layer node outputs. Simulation results using MATLAB/Simulink demonstrate that, under different test conditions, the system employing RBF neural network-PID control exhibits a smaller overshoot compared to traditional PID control systems. Furthermore, in steady-state circumstances, the settling time is shortened by around 1/2 and is decreased by roughly 1/4 in the presence of disturbance signals. These results unequivocally demonstrate that, in the presence of fluctuating test conditions, the new structure of this combination greatly improves the control precision and tracking performance of DC-DC converters.