Parameter identification of induction motors using stochastic optimization algorithms

In the area of optimization, applied research typically focuses on finding the best possible solution to the practical problem at hand. In contrast, a large part of basic research aims at developing novel algorithms with improved performance. In practical application, most studies employ rather simple algorithms. On the other hand, most novel algorithms are tested on a suite of artificial benchmark problems. Hence, there is a research gap between applied and basic research since the former focuses on the problems, whereas the latter concentrates on the algorithms. In this paper, we address the research gap for parameter identification of induction motors. In our experiments, we compare the performance of eight stochastic optimization algorithms on identification of two induction motors. The eight algorithms represent four main groups of algorithms presently used for numerical optimization. The four groups are: local search (LS), evolution strategies (ESs), generational evolutionary algorithms (EAs), and particle swarm optimizers (PSOs). The comparison includes a simple and an advanced algorithm from each group. From our experiments, we draw two conclusions. First, the advanced algorithms had significantly better performance compared with the simple algorithms. This underlines the importance of using advanced algorithms when approaching realworld problems. Furthermore, the improved performance justifies and motivates the development of more advanced techniques. Second, population-based stochastic optimization techniques (ESs, EAs, and PSOs) significantly outperformed the local search algorithms on both problems.