Font Size: a A A

Research On Particle Swarm Optimization For High-Dimensional And Multi-Objective Optimization Problems

Posted on:2015-09-06Degree:MasterType:Thesis
Country:ChinaCandidate:Z MaFull Text:PDF
GTID:2298330467987076Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
There are a lot of optimization problems in the real world, especially in the field of engineering practice and scientific research. As the industrial processes become complicated, more and more problems become high-dimensional, multi-objective. However, most of them are multimodal or disconnected that it is difficult to solve by the traditional optimizer. While the Evolutionary Algorithms (EAs) can solve these optimization problems effectively, more and more researchers pay attention to EAs. The aim of this article is to explore the theories and mechanisms of EAs and design effective algorithms and strategies for high-dimensional and multi-objective optimization.The main research works in this article consist of the following aspects:To balance the convergence speed and the diversity of Particle Swarm Algorithm, this paper proposed a Hierarchical Particle Swarm Optimizer with Random Social Cognition (HPSO-RSC). During the execution process of HPSO-RSC, the social environment is changed dynamically, and each particle is not only attracted by its previous best particle and the global best particle of the whole population, but also attracted by all other better particles randomly. During the early stage of the execution process, to speed up convergence of the algorithm, the particles are inclined to choose the global best particle as cognition object. On the other hand, during the late stage of the execution process, to keep the diversity of the population, the particles are inclined to choose the particles that better than themselves as cognition object. To solve the large scale global optimization problem, the algorithm is integrated into a cooperative coevolution framework with an efficient variable interaction checking method. Simulated experiments were conducted on the CEC2008benchmarks. The result demonstrates that, HPSO-RSC has strong ability to find the global optimum for most of the benchmark problems.In the evolution process of multi-objective optimization problems, the Pareto solution set becomes larger and larger. So the strategy that selects individuals from population randomly could hardly produce new better individual. To improve the proportion of getting better individuals by crossover operator, this paper proposed a Multi-Objective Particle Swarm Optimization based on Decomposition with Guided Crossover(MOPSO/D-GC).In MOPSO/D-GC, we check the correlation between the decision variables and the objective functions firstly, and finds the better gene segments corresponding with different objective functions. Then in the process of optimization, gene segments corresponding with different objective functions are token out to do guided crossover operation, and produce new individual. In order to get more heuristic from the neighborhood, we regard the historical improvement of other individuals in neighborhood as an individual’s learning factor. On the other hand, this paper proposed a dynamic scheduling approach that selects different individuals for next optimization according to their decrease of fitness, and checks its effect by experiment. Experimental results of MOPSO/D-GC algorithm on the unconstrained multi-objective test instances for CEC2009demonstrates that, the MOPSO/D-GC is able to find the relatively ideal Pareto optimal solutions. For most multi-objective test functions, the obtained Pareto front by MOPSO/D-GC is very close to the true Pareto front; but for a few test problems with multiple local Pareto front, the result is not so good.
Keywords/Search Tags:Particle Swarm Optimization, Large Scale Global Optimization, Multi-objective Optimization, MOEA/D, Evolution Algorithm
PDF Full Text Request
Related items