System-level Parameters Identification for DC-DC Converters Based on Artificial Neural Network Algorithm

Abstract

The existing Artificial Neural Network (ANN) methods for parameters identification are suffering from either training cost or low estimation accuracy. To overcome the aforementioned challenges, a novel parameters identification method based on ANN and electrothermal model is proposed in this paper. The proposed electrothermal model is more realistic than ideal model, including the circuit model of DC-DC converter, the thermal model of self-heating device and temperature feedback. Electrothermal model is subsequently used to train and generate the proposed ANN, which consists of two networks, ANN1 and ANN2. ANN1 is used to estimate the on-state resistance of MOSFET and ANN2 is used to estimate the capacitance and equivalent series resistance of capacitor. The proposed ANN method is validated on a 500W buck converter. The outcomes of this paper serve as a key step for achieving noninvasive, cost-effective, high-accuracy and multi-components parameters identification for DC-DC converters.