Performance characteristics of reference modification control DC-DC converter

Abstract

The purpose of this paper is to show performance characteristics of the reference modification control dc-dc converter which uses neural network and model controls. In the presented method, the neural network controller is used to modify the reference in the proportional control term of the conventional PID control. The neural network controller is repeatedly trained using former predicted data to predict the output voltage. After the training, the reference in the P control is modified by the predictor to reduce the difference of the output voltage and the desired one. This neural network control works to improve the transient response, however, it is difficult to improve transient response greatly when the operation mode across the discontinuous conduction mode and the continuous conduction mode. The model control is adopted simultaneously to ensure the performance from the no-load condition to the full-load condition as the model control is modified the bias term of the PID control in both steady and transient states. As the result, the convergence time of output voltage in the presented method is improved by 83% than the conventional PID control. Furthermore, the undershoot of output voltage is improved by 75% than the conventional PID control.