Data-Driven Control of DC-DC Power Converters Using Levenberg-Marquardt Backpropagation Algorithm

Abstract

The majority of the controllers are designed around linearized small signal models of switching power converters. These models often encounter shortfalls in capturing the dynamics and underlying behaviours of the switching converters. Hence, in order to comply with the stringent requirement for voltage regulation in many modern applications which are plagued by non-idealities such as load disturbance and varying parameters, the use of adaptive, nonlinear and intelligent controllers becomes pivotal. It is against this backdrop that this paper proposes a data driven control using a four-layered feedforward neural network controller which is able to achieve a near-optimal performance in the output waveforms of a synchronous dc-dc buck converter. The training data for the neural network are extracted from the simulation of the converter using the designed type II compensator in current mode control with load current feedforward, considering wide range of dynamic changes in load current and input voltage. Results clearly show that the proposed ANN controller gives better performance than the conventional Type-II and Type-III compensators.