Genetic Algorithm And Its Application To Water Problems

On the basis of the systematic and deep discussion and study on geneticalgorithm(GA),in this dissertation, an improved simple geneticalgorithm(SGA),named accelerating genetic algorithm(AGA),is presented, and AGAis applied systematically to water problems such as flood disasterevaluation ,constructing mathematics models for hydrology and water resources ,andoptimization problems in water environment , which establishes a system closelycombining theoretical research with practical application. The major contents aresummarized as following:1.Some shortcomings of the traditional optimization methods for resolvingcomplex water problems are pointed out, and the origination, development andapplication of genetic algorithm are stated and summarized. A basic approach tostudying genetic algorithm is presented ,which is that the process of studying geneticalgorithm is the same as the process of the evolution of genetic algorithm ,the sameapplies to the processes of selecting strategy of scientific research, evaluatingachievements in scientific research, understanding and remaking natural.2.The intrinsic contradiction between the total optimization and the convergenceof SGA is pointed out.3.On the basis of the above summarization, eleven approaches to improvementon SGA are given .AGA is presented for accelerating the evolution of SGA, taking account of thetotal optimization of SGA. The convenient configuration of the control parameters ofAGA is obtained ,that is:(1)the encoding length of bit string(e) ,the probability ofcrossover(pc) and the probability of the mutation of string(pm) can be chosen 10,1.0and 1.0 respectively ;(2)the population dimension(n) and the number of superiorindividuals produced in the process of every evolutionary iteration(s) have theexperimental relation as following: s/n>n/(e·2e), and the value of n is suggested to bechosen as 300 or larger than 300,so the value of s is suggested to be chosen as 10 orlarger than 10 when AGA is applied to the practical problems;(3)only twoevolutionary iterations of SGA are done in each acceleration evolutionary iteration.The theoretic analysis and the testing result of AGA under eleven differenceoptimization problems show that the characteristics of AGA are good and its field ofapplication is very wide.4.A new algorithm for training BP artificial neural networks ,named BP-AGA...