Machine-Learning-Based Parameterization of Adaptive Notch Filters for CM Noise Reduction in Motor Inverters

Fast-switching semiconductor components are increasingly used in motor inverters for drive systems. The short rise times and high voltage levels improve the efficiency of the devices but lead to an increase in electromagnetic emissions. The conducted emissions are typically reduced with filters. Active EMI filters (AEF) promise weight and volume reduction compared to passive filters. Digital AEF (DAEF) concepts offer noise suppression at higher frequencies, but adjustment of a large number of filter parameters can be challenging. The dependence of the many parameters on each other and the strong non-linearity of the DAEF often lead to unexpected behavior. In this paper, a parameter optimization algorithm based on supervised learning is proposed for the design of a special kind of DAEF. The aim is to find the optimal parameters for several adaptive notch DAEFs to reduce EMI noise in a wide frequency range. The new approach is implemented and tested for applicability in a simplified laboratory setup.