Evaluating hybrid optimization algorithms for design of a permanent magnet generator

Abstract

Particle Swarm Optimization (PSO) and Genetic Algorithms (GA) are used to minimize the cost of a permanent magnet (PM) synchronous generator with concentrated windings for tidal power applications. With the use of MATLABs global optimization toolbox, it is possible to run several optimization algorithms on the same problem, and to combine the two stochastic solvers GA and PSO with the gradient based solver fmincon to produce two hybrid optimization solvers. It has been shown that a complex machine design problem with tight constraints and a narrow solution space is difficult to solve for both a GA and for PSO. Both GA and PSO were unable to find the optimal value on their own. Hybrid versions of GA and PSO gave better results. The average minimum costs found with hybrid PSO and hybrid GA were 1.07 and 1.11 times the global minimum. When the integer value was set to the optimal value, the hybrid GA found a mean cost only 1.01 times the global minimum. For both algorithms, it was necessary to increase the population size to improve the fitness functions and reduce the variance.