Modeling loaded starter motor with neural network

In this paper a three-layered, feedforward neural network based model of a starter motor was introduced. Teaching and validating datasets are collected from real system measurements where different character of load torque was applied on the motor's shaft. Different types of training datasets were used to investigate its influence on the trained network. Beside the well-known MSE, other information criteria like AIC, BIC, FPE were applied to reduce the time consumption of the training process and also to analyze its influence on the resulting model. To achieve the best result, the structure of the neural network was also changed.