Steady-State Torque Prediction of Inverter-Fed Induction Machine Based on Multi-Layer Perceptron Neural Network

Overall performance estimation of the inverter-fed induction machine by finite element analysis is a very time-consuming task, which is result of the long transient response time and various operating conditions of induction machines. To solve this problem, a method based on Multi-Layer Perceptron neural network is proposed. Firstly, minimum amount of operating conditions are calculated by finite element analysis to obtain the steady-state torque. Then the Multi-Layer Perceptron method is adopted to process the calculated result and build the prediction branching models classified by slip frequency. Finally, the prediction branching models are merged to construct the complete prediction model. It is tested that the steady-state torque obtained by this method has high accuracy. Compared the traditional method that requires sweeping all operating conditions, it has greatly reduced the time cost and operation cost of data acquisition.