Font Size: a A A

Research And Application Of The Simulated Annealing Particle Swarm Mixing Algorithm

Posted on:2011-07-25Degree:MasterType:Thesis
Country:ChinaCandidate:L J LinFull Text:PDF
GTID:2178360308465539Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
Since the 1980s, swarm intelligence has caused concerns of many researchers as a new field,and has become the hot spots and cutting-edge of interdisciplinary fields of social,economic,biological and etal. Artificial neural networks,simulated annealing,genetic algorithms,particle swarm optimization and ant colony algorithm developed by simulating certain natural phenomena and processes providing a new way of thinking and means for the optimization theory. PSO algorithm proposed in 1995 is simple,easy to implement,not many the parameters to be adjusted and faster convergence. It has been widely used in the objective function optimization,dynamic environment optimization,neural network training and many other areas,and it has become an independent research branch in IEEE Annual Conference Of Evolutionary Computation.Simulated annealing algorithm was proposed by S.Kirkpatrick and etal in 1982 as an extension of local searching algorithm,which is a stochastic optimization method established by simulating annealing mechanism of metal. Accepting the new model of SA make it into a global optimal algorithm, and have obtained theoretical proof and verification of practical application. Because of the existence of such advantages, it is introduced successfully the field of the combinatorial optimization theory.In recent years,the algorithm caused extensive attention in the field of large-scale optimization design,numerical analysis,complex layout and etal.Computer technology,multimedia technology and the rapid development of Intemet technology lead to the emergence of a large number of images.A very important but challenging research topic at present: how to retrieve the images needed effectively and quickly from large-scale image databases. Content-based image retrieval technique solve the problem which retrieve relevant images from image database using the automatic acquisition of image features. In recent years, research of this technology is very active and have applications in many fields.This article combined simulated annealing algorithm and particle swarm optimization algorithm,and carried out some exploratory research focusing on a number of key technologies of content-based image retrieval. The contents of the paper belongs to the research focus of the image information retrieval and intelligent algorithm optimize field and have considerable theoretical and practical value, to provide support platform for the design of the idea of new intelligent classification retrieval. Its main tasks are: 1. Propose a Dynamic Self-Adaptive Particle Swarm Optimization(DAPSO)PSO algorithm is apt to fall into local optimum, and exists the problem of premature convergence. Many studies have focused on the improvement of the inertia weight w .With different inertia weight particle fulfilling their duties and global optimization and local optimization performing at the same time, the algorithm can guarantee a very good compromise between the global convergence and the convergence rate. When the algorithm does not search the global best fitness value, or does not meet the optimal requirements, we can use inertia weight variation strategy. We can generate a small magnitude of the disturbance by larger probability in order to achieve local search,also can have a significant disturbance in order to achieve the long migration to step out of the local minimum area.2. Propose a Dynamic Self-Adaptive Particle Swarm Optimization Model based on the "Small World"(DWPSO)Existing PSO algorithm and its variants all have slow convergence and easy local extremum problem, and the dynamic self-adaptive particle swarm algorithm have a premature convergence, then this paper presents a Dynamic self-Adaptation of Particle Swarm Optimization based on "Small World". On the dynamic micro-world particle swarm algorithm foundation,DWPSO introduced cross and variation mechanism which can reduce computing time as well as avoiding the precocious phenomenon.3. Combined particle swarm optimization algorithm and simulated annealing algorithm, propose SA-DWPSO hybrid algorithm.It has proved that PSO algorithm can not guarantee converge to the optimal solution, or even local optimal solution in theory.Simulated annealing algorithm has been proven to converge to the global optimal solution set by probability one, so we can use the simulated annealing algorithm as the convergence basis of the PSO algorithm. The combination of the local convergence of PSO and the global convergence of simulated annealing effectively overcome the premature convergence of particle swarm algorithm and accelerate the convergence speed.4. Simulated annealing algorithm and improved particle swarm algorithm are combined cooperatively search.The search can keep both the advantages and have good complementary. The hybrid algorithm is applied to feature-based image retrieval and it achieve better classification results. Using VC + +. NET 2008, SQL Server2005 database system and MATLAB has developed completely in Windows XP platform. Perform experimental analysis respectively for image retrieval,classification,optimization and etal.Design results is satisfactoried.
Keywords/Search Tags:Particle Swarm Optimization, Simulated Annealing Algorithm, Gaussian Mutation, SA-DWPSO, Image Classification
PDF Full Text Request
Related items