Prediction of Motor Temperature using Linear Regression

The direct measurement of the temperature of a permanent magnet synchronous motor (PMSM) is difficult due to its complexity of the construction of the motor. It is difficult to embed thermal sensors in the motor. Thus, the temperature of various components of the motor must be modeled from other parameters. The classical methods of thermal modeling lack accuracy and require expertise on heat models as well as knowledge of the individual motor construction. In this paper, efficient and fast predictive linear models are evaluated. Linear regression with gradient descent and normal equations is evaluated to predict the dynamic temperatures inside PMSM. The features used to train the data are selected as per correlation analysis. Results are further optimized using regularization techniques such as L1 and L2 regularization. K-nearest neighbor regression is evaluated, and then different predictive models are compared.