Parallel genetic algorithm and its application to linear synchronous motor optimization
Liudmyla Lysenko, V.I. Omelyanenko, S.A. Sergeev · International Journal of Applied Electromagnetics and Mechanics · 1998
The problem of globally optimizing linear synchronous motor (LSM) design parameters in order to minimize power consumption is of great practical interest to those involved in engineering MAGLEV trains. The paper describes an optimization technique based on a Parallel Genetic Algorithm in which separate subpopulations of possible solutions in the search space are interpreted not as demes but as individuals with the subpopulation-points co-ordinate matrices identified as the individual chromosome sets. The second exclusive detail of the algorithm results from using a deformed polyhedron method as a means of improving the algorithm ability to fine-tune the solution. The subpopulation points are treated as polyhedron apices, after which search heuristics borrowed from the Nelder–Mead method are applied to polyhedrons to provide their adaptation. The developed algorithm was examined via a ten-dimensional multiextrema test function before being applied to the multicriteria problem of LSM optimization. In both cases, it has coped with the tasks quite successfully, and has improved on results reported earlier. The paper also explores the lower bound of a feasible domain in the LSM criteria space.