SVM-based and Classical MRAS for On-line Rotor Resistance Estimation: A Comparative Study

Abstract

This paper makes a comparison between the performance of a classical model reference adaptive system (MRAS)-based observer to estimate the rotor resistance of the SCIM and the performance of a support vector machines (SVM)-based MRAS observer to estimate that parameter. The most important parameter of the squirrel cage induction motor to be considered in indirect vector control is the rotor resistance; because of this parameter has a strong influence in the performance of the drive. It is well known, if there is a mismatching between rotor resistance of the machine (varying with temperature, saturation, skin effect) and its corresponding one in the controller (fixed), the latter cannot determine the correct position of the synchronous d-q axes and the consequence is the lost of the field orientation. The complete drive system including a time-varying rotor resistance model for the SCIM was simulated. Results showed the performance of the SVM-based estimator was better than performance of the classical MRAS-based estimator for the same operation conditions of the drive system. This work showed the powerful of the SVM used as regressor to estimate an unknown and inaccessible rotor resistance parameter of the SCIM, which demonstrated this new artificial intelligent branch has a promissory future to solve many different problems in engineering field applications.