Simulation and analysis of neural network-based induction motor control system

The direct torque control of induction motors has gained popularity in industrial applications mainly due to its simple control structure. The stator voltage vector could either be sensed from the machine terminals or reconstructed using inverter switching status and the DC bus voltage. A novel method of parameter identification using wavelet neural network is presented for performance improvement in low speed status of induction motor system. The advantage of the wavelet logarithmic time frequency bands is in achieving flexible frequency resolution, thus making it able to extract both high-frequency and low-frequency components from the original signal. In training process, the wavelet network learns adequate decision functions and arbitrarily complex decision regions defined by the weight coefficients. The accurate stator flux vector and electromagnetic torque are acquired by the network output once the instants are detected, where the direct torque control can be applicable in the low region and the inverter control strategy can be optimized. The simulation results show that the proposed method can efficiently reduce the torque ripple and current ripple, acquiring good performance in low speed state.