Speed Estimation of Induction Motor Using Gaussian Process Regression

The control of an induction motor (IM) drive is a complex process and requires speed estimation, which is dependent on various machine parameters. The linear regression approach reduces this dependency by eliminating the need for flux computation and gain adjustment of the PI controllers. The performance of the linear regression model deteriorates when data is noisy, the best fit deviates from the desired value. The Gaussian Process (GP) model is a non-parametric model, that can incorporate these noisy measurements and model uncertainties. In this paper, the speed of the IM drive with vector control is estimated using Gaussian Process Regression (GPR). The fictitious quantity
$X$
= ū*×
$I$
is used to eliminate the calculation of flux in the stator or rotor and its characteristic of stable drive in all four quadrants. The GP is viewed as a surrogate model, the prediction distribution obtained gives a confidence interval, used for usefulness validation that also distinguishes GP model from other black-box models. Furthermore, the Bayesian approach of GPR involves lesser complex evaluations making this approach simple and agile.