Research On The Particle Swarm Optimization In Dynamic Environment

The world is dynamic. Every day, in every aspect of the human experience, conditions, requirements, and goals change. A secretary works hard to schedule a meeting between various parties, only to find at the last minute that a fresh conflict has arisen, requiring that the schedule be completely reworked. Particle Swarm Optimization (PSO) is a population-based, self-adaptive search optimization technique that has been applied to find optimal or near-optimal solutions for real-world optimization problems, the optimization in dynamic environment brings new challenges to PSO even to evolutionary computation as a whole. Carlisle in2001proposed an improved APSO. In the new APSO algorithm, one particle, call a "sentry", was chosen as the environment. The particles that receive the sentry’s alert will reset their memory and replace the personal best vector as well as the fitness value with the vector of particle’s current location and the new fitness value, respectively. However, in an environment with that exhibits local change phenomena, it is possible that the fixed sentry cannot detect these changes. So we need research new optimization algorithm which follow the principle of swarm intelligence and quickly respond to changing environment.Because of PSO’fast convergence and easy to realize and particle’ flight mechanism, PSO is very suitable for dynamic environment. This work analyzes why one such static problem algorithm, the PSO, fails when the environment is dynamic, and what modifications to the algorithm are necessary to overcome this deficiency.The innovations of the thesis are described as followings:(1) In this thesis, rank-based selection is proposed for the particle swarm optimization, and the thesis applies mutation along with rank-based selection to improve the diversity of the population. Time-varying values are used for the acceleration coefficients with both methods to keep a higher degree of global search and a lower degree of local search at the beginning stages of the search.(2) In order to achieve the optimal solution in dynamic environments and prevent particles from being over-crowed, this thesis proposes hierarchical multi-population PSO, the entire population is divided into two hierarchical populations. One is for the ordinary population and the other population, termed dynamic population, is designed for automatically tracking various changes quickly in dynamic system. The dynamic population is separated further into several groups.(3) PSO based on variable recognition is inspired by the renewal of knowledge in human society. In the algorithm, we give a new parameter, the recognition variable coefficient. The recognition value will decrease at the rate of the recognition variable coefficient over time.(4) In this study, a novel computational framework based on cultural algorithm has been proposed using knowledge stored in the belief space to re-diversify the population right after a change takes place in the dynamic of the problem. The population space (PSO) will be initialized and then divided into several swarms according to the closeness of the particles. The belief space is then initialized. Next we apply acceptance function to select some particles which will be later adopted for the belief space. Belief space consists of five sections, situational, history, domain, normative and spatial knowledge. Next we apply influence functions to the belief space in order to select the key parameters of PSO for next iteration, including the repulsion factor for each particle, personal best, swarm best and global best. Through a scheme using information from a belief space, the change in dynamic will be detected. As soon as the change is detected, influence function applies to the belief space to perform the repulsive diversity-promoted migration among swarms.(5) We apply the regional reconnaissance method to PSO. It leads the particles to extend their search area. Thus, not only the precision of solution but also the time consumed is improved when enl environment change. Conventional PSO usually leads the particles go into the wrong direction of evolution. To resolve this drawback, when particles response to change, we spread some particles into ambiguous solution space.