Particle swarm and simulated annealing for multi-local optimization

Abstract

Particle swarm and simulated annealing optimization algorithms proved to be valid in finding a global optimum in the bound constrained optimization context. However, their original versions can only detect one global optimum even if the problem has more than one solution. In this paper we propose modifications to both algorithms. In the particle swarm optimization algorithm we introduce gradient information to enable the computation of all the global and local optima. The simulated annealing algorithm is combined with a stretching technique to be able to compute all global optima. The numerical experiments carried out with a set of well-known test problems illustrate the effectiveness of the proposed algorithms.