Visualization of Multi-Objective Switched Reluctance Machine Optimization at Multiple Operating Conditions with t-SNE
Shen Zhang,Shibo Zhang,Sufei Li,Liang Du,Thomas G. Habetler
DOI: https://doi.org/10.48550/arXiv.1911.01024
IF: 5.414
2019-11-04
Machine Learning
Abstract:The optimization of electric machines at multiple operating points is crucial for applications that require frequent changes on speeds and loads, such as the electric vehicles, to strive for the machine optimal performance across the entire driving cycle. However, the number of objectives that would need to be optimized would significantly increase with the number of operating points considered in the optimization, thus posting a potential problem in regards to the visualization techniques currently in use, such as in the scatter plots of Pareto fronts, the parallel coordinates, and in the principal component analysis (PCA), inhibiting their ability to provide machine designers with intuitive and informative visualizations of all of the design candidates and their ability to pick a few for further fine-tuning with performance verification. Therefore, this paper proposes the utilization of t-distributed stochastic neighbor embedding (t-SNE) to visualize all of the optimization objectives of various electric machines design candidates with various operating conditions, which constitute a high-dimensional set of data that would lie on several different, but related, low-dimensional manifolds. Finally, two case studies of switched reluctance machines (SRM) are presented to illustrate the superiority of then t-SNE when compared to traditional visualization techniques used in electric machine optimizations.