A backpropagation neural network controller trained using PID for digitally-controlled DC-DC switching converters

Abstract

In this paper, a backpropagation (BP) neural network controller trained using PID for digitally-controlled DC-DC switching converters is proposed to improve the transient response performance. To reduce the hardware cost and power consumption, only one BP neural network controller is used in the proposed controller, which is trained to learn the PID control algorithm and obtain the optimal control coefficients to fit the input-output relationship adaptively under different operating points. Furthermore, the proposed controller uses a single-hidden-layer BP neural network to reduce the computing time. Then, a buck DC-DC switching converter with the proposed controller is designed and realized on the field programmable gate array and printed circuit board. The experimental results indicated that it yields a better transient performance in a DC-DC switching converter than does a conventional neural network PID controller that contains two sub-controllers: the settling time is improved by at least 50% and the hardware resources required for a PID compensator are saved.