0-7803-7488-6/02/$17.00 ©2002 IEEE. 1215 Adaptive Particle Swarm Optimization on Individual Level Xiao-Feng Xie, Wen-Jun Zhang, Zhi-Lian Yang Institute of Microelectronics, Tsinghua University, Beijing 100084, P.R.China Email: xiexiaofeng@tsinghua.org.cn Abstract - An adaptive particle swarm optimization (PSO) on individual level is presented. By analyzing the social model of PSO, a replacing criterion based on the diversity of fitness between current particle and the best historical experience is introduced to maintain the social attribution of swarm adaptively by taking off inactive particles. The testing of three benchmark functions indicates it improves the average performance effectively. Key words: adaptive particle swarm optimization, evolutionary computation, social model 1. Introduction Modern heuristic algorithms are considered as practical tools for nonlinear optimization problems, which do not require that the objective function to be differentiable or be continuous. The particle swarm optimization (PSO) algorithm [1] is an evolutionary computation technique, which is inspired by social behavior of swarms. It has been used for approaches that can be used across a wide range of applications, as well as for specific applications focused on a specific requirement [2]. Work presented in [3] describes the complex task of parameter selection in the PSO model. PSO has been proved to be a competitor to the standard genetic algorithm (GA). Comparisons between PSO and GA were done with regards to performance by Angeline [4], which points out that the PSO performs well in the early iterations, but has problems in reaching a near optimal solution in several benchmark functions. To overcome this problem, some researchers have employed methods with adaptive parameters [5, 6]. One is deterministic mode, which the PSO parameters are changed according to the deterministic rules, such as a linear decreased inertia weight as the number of generation increasing [5], which are obtained according to the experience. The other is adaptive mode, which adjusts the parameters according to the feedback information, such as fuzzy adaptive inertia weight [6]. However, currently, we still have not captured the relations between different parameters and their effects toward different problem instances, such as dynamic optimization problems [2], due to the complex relationship among parameters. Unlike the former efforts on adjusting PSO parameters, this paper will propose an efficient approach by adapting the swarm on individual level, which is realized by replacing the inactive particle with a fresh one in order to maintain the social attribution of swarm, according to the analyzing for the model of PSO. Both standard and adaptive versions are compared on three benchmark problems. The results suggest that the adaptive PSO enhance the performance effectively. 2. Standard particle swarm optimization (SPSO) PSO is similar to the other evolutionary algorithms in that the system is initialized with a population of random solutions. Each potential solution, call particles, flies in the D-dimensional problem space with a velocity which is dynamically adjusted according to the flying experiences of its own and its colleagues. The location of the ith particle is represented as X i = (x i1 ,…, International Conference on Signal Processing (ICSP), Beijing, China, 2002: 1215-1218 Related Papers & Source codes: http://www.adaptivebox.net/research/fields/algorithm/pso/index.html