Study On The Computation Performance And Application Of The Coarse-grained Parallel Genetic Algorithms

Genetic algorithms(GAs) are optimization algorithms developed by integrating genetic evolution rules and stochastic optimization theory.The CPGA is an important improved form of GAs.It has better performance than the classic GA.It can solve the contradiction of premature convergence and slowness of local convergence.The paper mainly discusses CPGA.As a kind of heuristic searching algorithms,the computation results are unstable and unrepetitive.In the present,few works are available for the process mechanisms of genetic operations,thus many assumptions cannot be proved mathematically.The computation process of GAs will be affected by kinds of stochastic perturbations.On the other hand, many examples show that the results of GAs have certain reliability in statistical sense.The paper improves the reliability of GAs by the method of taking average value of reiteratively computation.The variance of multi-computations results and the average value of these results are taken to be two criterions to estimate the stability and convergence of the computations,and analysis on the computation processes are proceded basic on the criterions.The paper valuates the computation efficiency of GAs by comparing the computing times of GAs and globle stochastic searching algorithm.The main devotion of this work is the computations on the classic test functions that have different mathematical characteristics.According to the comparison analysis,the computational characteristics and the optimal parameter setting results are obtained.The paper has some conclusions on the stability and reliability of GAs by analyzing the results of repeating computing statistically.We also get to know that better computation performence can be gotten by using GAs and their improvement when we solve practical large-scale engineering optimization problems.The paper makes some improvements to mutation operation.The improved methods basic on three considerations,and the paper can realize each one with different ways. There are six kinds of improved methods.The paper analyses their computation characteristics,and get the results that stochastic generating new individuals in every certain evolution generations is optimal way,and the computation results of the six improved methods are better than the ones obtained by classic GA without any improvements.In the paper implement,classic GA is influenced by the problems of premature convergence and convergent speed.CPGA is an important improvement of classic GA, and can solve the problems mentioned here.In the paper,CPGA are used to do examples analysis corresponding to the different parameters of the number of subpopulations,the size of the population and the evolution generations,the results on computation efficiency and practical influences are obtained.And by comparing the results of classic GA,the paper demonstrates that better results and function process can be obtained by using CPGA,which has higher population diversity and better computation stability.The paper systematically describes the synchronous migratory and asynchronous migratory modes,and does example analysis to the performance of synchronous migratory and asynchronous migratory with different parameter setting.The results show that: synchronous migratory as a special case of asynchronous migratory,it is neither the best nor the worst form of asynchronous migratory.In order to get the best computation results, the parameters of the basic migratory space and migratory probability must be well set; compared with multi-modality problems,it is better to take synchronous migratory mode for unimodality problems.The paper also demonstrates a kind of improved CPGA of which each subpopulation uses different parameter settings.The improved methods take use of the frameworks of coarse-grained parallel evolutions,and make the results that CPGA can be find different local optimization solutions for different subpopulations searching.This method improves the subpopulations diversity and computation performance of CPGA.The paper does computation tests to find that the method is effective,especial for the multi-modality problems.And the paper can avoid the computational limitations in solving practical engineering optimization problems,which provides convenience in solving the problems.The paper solves the problem of short-term plan of the step power plants on the LanCangJiang by CPGA.In this article,we build models,transform the models to adjust the characteristics of GAs,and then get optimization solutions by CPGA,which is set by known evolution strategies,and analyze the computation results.By solving the practical engineering problems,it can be concluded that CPGA can be well applied in practical problems.The practices indicate that GAs have advantages especially for the models that will be computed many times by different initialization data.Using the results mentioned above,better computation results can be obtained,when the paper solves the practical engineering optimization problems.In the paper,the improvements to the classic GA provide theory of strategy setting methods for the practical large-scale engineering computations,which is a fine application of GAs.In theoretic aspect,the researches to the example analysis and parameter setting provide great supports to the principle studies of GAs.