Electromagnetic emission measurement prediction of buck-boost converter circuits using machine learning methods

Abstract

In this paper, a prediction system has been developed using machine learning techniques to obtain the conduction emission levels ensure they remain below the limit values specified in test standards. An LED (Light-emitting diode) driver circuit based on a buck-boost type DC-DC converter has been employed in the experiments. Standards-compliant conducted emission testing processes have been performed and measurement results have been used to generate datasets. These datasets have been organized and processed according to the targeted machine learning methods. GPR, has achieved the highest success rate of 99% among ANN and regression methods. In order to improve the performance in EMI harmonic prediction, training was conducted using deep learning, and the obtained model has a mean squared error of 0.78. The harmonics are well captured with the method and the results are in good agreement with measurements. Consequently, the number of required pre-compatibility tests for a similar topology can be significantly reduced.