Chinese Journal of Computers   Full Text
Title	The Statistical Analyses for Computational Performance of the Genetic Algorithms
Authors	YUE Qin FENG Shan
Address	(Department of Control Science and Engineering, Huazhong University of Science and Technology, Wuhan 430074)
Year	2009
Issue	No.12(2389—2392)
Abstract &   Background	Abstract In this paper, the authors discuss the reliability of the GAs by reiteratively computing the multi-dimensional analytic functions and statistical analysis of the results. The analysis results show that the GAs have certain stability; it could improve the reliability by reiteratively computation and estimates the effects of improvements. Keywords genetic algorithms; computational stability; confidence interval Background Genetic algorithms (GAs) are optimization algorithms developed by integrating genetic evolution rules and stochastic optimization theory. As a kind of heuristic searching algorithms, the computation results are unstable and unrepetitive; the evolution process has directional randomicity, which increase the general average of the fitness value of population. In the present, few works are available for the process mechanisms of genetic operations, many assumptions can not be proved mathematically. The computation process of GAs will be affected by kinds of stochastic perturbations such as random initial population and stochastic mutation operation, especially affected by the random initial population. On the other hand, many examples show that the results of GAs have certain reliability in statistical sense, which inspires us improve the reliability of GAs by the method of taking average value of reiteratively computation. Heuristic searching algorithms including GAs are usually used to solve the large-scale complex optimization problems, which could not be done by traditional optimization arithmetic. The results computed by GAs to these problems may not reach the exact solution either, which bring difficulties in evaluations for practical engineering optimization computation results. Many forms of GAs are used in practical. The classical GA is the simplest form, but it is lack of practicality. This paper uses the coarse-grained parallel genetic algorithm(CPGA), which is a important improved form of GAs. The coarse-grained parallel genetic algorithm has better performance than the classic GA. The CPGA can solve the contradiction of premature convergence and slowness of local convergence. 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.