Comparison of adaptive Neuro-Fuzzy-based space-vector modulation for two-level inverter

Space Vector Modulation (SVM) is an optimal pulse width modulation technique for an inverter used in variable frequency drive applications. This paper proposes a Neuro-Fuzzy based Space Vector Modulation (SVM) technique for voltage source inverter and its performance is compared with the conventional based SVM and Neural Network based SVM methods. This scheme is five-layer network, receives the d-axis and q-axis voltages information at the input side and generates the duty ratios as an output for the inverter circuit. The training data for Neural Network and adaptive Neuro-Fuzzy is generated by simulating the conventional SVM. Neuro-Fuzzy uses the hybrid learning algorithm for training the network. Due to this learning algorithm, the required training error can be obtained with less number of iterations compared to Neural Network. The simulation results obtained are verified experimentally using a DSPACE kit (DS1104). The simulation and experimental waveforms of inverter line–line voltages at different switching frequencies is presented. The Total Harmonic Distortion (THD) of line–line voltage with Neuro-Fuzzy, Neural Network and conventional based SVM methods for various switching frequencies are presented.

► Inverter performance with conventional, neural and ANFIS based SVM is compared.
► The simulation results evaluated experimentally using DSPACE kit.
► The training time of the ANFIS is compared Neural Network based SVM.
► The performance of inverter at two different switching frequencies is analyzed.
► Induction motor performance is also studied under different operating conditions.