Multi-Phased And Diversity-Maintianed QPSO Algorithms And Applications In System Identification

A novel class of evolutionary computation technique-Swarm Intelligence Algorithm is discussed, among which the Quantum-behaved Particle Swarm Optimization (QPSO) algorithm is a recently proposed approach and is a variant of original Particle Swarm Optimization (PSO). QPSO is global convergent and will be a promising solver for complex optimization problem, which is shown by some previous work. Thus, the research of this paper will be of somewhat significance in evolutionary computation area.The traditional evolutionary algorithm called Genetic Algorithm (GA) is firstly formulated in particular and in turn the Swarm Intelligence Algorithm including PSO and QPSO. As we known, the most complained problem about evolutionary computation is Premature Convergence (PC), which is also inevitable in QPSO. Thus this work focuses on how to overcome it and consequently propose two enhanced QPSO: Multi-phased QPSO (MQPSO) and Diversity-maintained QPSO.In MQPSO, an approach of using multi-phase for the swarm is proposed to enhance the global search ability of the QPSO algorithm. When implementing the MQPSO, two phases are defined: convergence phase and explosion phase; the swarm is divided into two sub-swarms and set each one in different phase. And therefore the whole particle swarm could undertake persistent search in the solution space, leading to an algorithm that can avoid premature convergence effectively.In our proposed DMQPSO, The Euclidian distance-to-average-point measure of the particles'personal best positions is used as diversity and a low bound value is set for the diversity measure to prevent the swarm from clustering. That means the diversity will be maintained higher than the low bound. When the diversity value declines to below the low bound, mutation operation is exerted on a randomly selected particle's pbest position until the diversity value returns to above the low bound. The experiment results on several benchmark function show that the diversity maintenance method may be a good technique to avoid premature convergence and may result in performance improvement of the QPSO in many cases.The applicability of QPSO is also explored to system identification problems. The system identification problem is key to optimal control. It can be reduced to an optimization problem.PSO and QPSO are used to test some benchmark system identification problems. The simulation results show that QPSO could generate better solutions for parameters of the tested systems with better fitness function value. It has also been shown that QPSO has faster convergence rate than PSO on system identification problems.