Dissertation > Excellent graduate degree dissertation topics show

Function Optimization PSO clustering algorithm

Author: ZhuShiJuan
Tutor: ZhuQingBao
School: Nanjing Normal University
Course: Applied Computer Technology
Keywords: Continuous Optimization PSO Grouping Stratified Vidor High modulus function Dynamic Optimization Diversity
CLC: TP301.6
Type: Master's thesis
Year: 2011
Downloads: 81
Quote: 0
Read: Download Dissertation

Abstract

Continuous optimization is widely used in all walks of life, has been a research focus. In the continuous domain, when solving the problem of the objective function or parameters do not change over time, it is called a static environment optimization, otherwise known as Dynamic Environment. To solve optimization problems such traditional optimization methods are based on mathematical solution gradient information, and requires the objective function is continuous and differentiable. This computing architecture were difficult to overcome the limitations, not extensive and accurate solution to achieve the objective function. For this reason, people try to use heuristic algorithms based on swarm intelligence to solve such problems, in which particle swarm optimization (PSO) is a simulation of natural biological behavior of a group of bionic intelligent algorithm that is optimized for the calculation of uncertain data, while PSO because of ideological intuitive, simple, small and tunable parameters and high efficiency of the implementation has been successfully applied to continuous optimization problems. But the standard PSO algorithm in solving high-dimensional multi-modal continuous domain function exists when trapped in local minima, slow convergence and other issues. To this end, we propose a solution space between partitions based on particle swarm optimization clustering algorithm, which will be the solution space is divided into several sub-spaces, each space is randomly assigned a set of particles, each set of particles in each sub-space for independent Search, which will be broken down into a single multi-extremal problems extremal problem, at least significantly reduce the need for optimization of the extreme number. In order to ensure that each species diversity and ergodicity search, introducing chaotic sequence to determine the initial position of the particle in each group. Experiments show that: the complexity of the algorithm for solving high-dimensional multi-mode function, high precision, fast convergence, the effect is satisfactory. Using standard PSO algorithm to solve dynamic continuous function optimization problem, because the solution process is to gradually converge to the optimal solution, in the later stage of evolution loses diversity of the population can not adapt to changes in the environment. Most of them are already computationally intensive algorithm, defects and poor adaptability. To this end, we propose a new grouping hierarchical dynamic particle swarm optimization. Taking into account the goal of optimization is to track changes in the extreme point, the optimal solution dynamic change on a global scale, the partition on the basis of the packet, each group of particles having a global search for the optimal implementation of the global search swarm, and the resulting results back to each group to guide the particles in each group at the grassroots level for local search, setting the partitions, each set of particles narrow your search and executed in parallel, greatly improving the efficiency of the algorithm. And in order to improve the algorithm's ability to adapt to the dynamic environment is proposed based on the set of optimal particle own information as well as the changing environment in each group detection method based on global best particle and other particles determine the average distance of the diversity of the environmental response methods, making algorithm can have a choice, with the direction of updating of the position of the particle groups. Experiments show that: the algorithm can quickly adapt to environmental changes, tracking dynamic extreme points, the effect is very satisfactory.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > General issues > Theories, methods > Algorithm Theory
© 2012 www.DissertationTopic.Net Mobile