Reduce common-mode voltage in cascaded multilevel inverter based on artificial neural network

Common-mode voltage due to different modulation in power converters has introduced numerous problems in electrical systems. This paper reviewed a Space Vector PWM (SVM), which can get rid of the common-mode voltage in multilevel inverters with fundamental switching frequency. But for difficulty of implement SVM algorithm, which requires complex and time-consuming online computation, a feedforward artificial neural network (ANN) with four-layer configuration is proposed, which receives angle command and modulation index as input and generate desirable PWM pattern for the three-phase cascaded multilevel inverter. The data derived from simulating of SVM-based scheme is employed to train the proposed ANN with MATLAB/Neural Network Toolbox offline. The performance of the trained ANN-based modulator is investigated in the 11-level cascaded inverters prototype, and the results of the experiment appear to be wonderful that common-mode voltage is reduced drastically and the power semiconductors operate with fundamental switching frequency.