Evaluation of Automatic Adjustment Methods for Voltage Reference to Compensate Dead-Time Distortion Using Machine Learning

In many power converters, a compensated voltage reference added to the voltage references to compensate for dead-time distortion is preset at the design stage. However, the compensated reference may deviate from the proper value when semiconductor devices are replaced or when device characteristics change due to the operating environment. Therefore, there are demands for active adjustment of the compensated reference. This paper evaluates and compares the effectiveness of three methods for adjusting the compensated reference using machine learning: (i) a method using the image classification model, (ii) a method using waveform feature-based linear regression model, and (iii) a method using one-dimensional CNN regression model with time series data. It is shown that the method (iii) is more effective than the others.