Schema Communicating Multi-Population Genetic Algorithms And Its Application In ANN Evolutionary Modeling

Aming at the problem of premature convergence in GA and the method of ANN evolutionary modeling , this thesis focues on the dynamic properties of multi-modal function optimization problem, schema communicating multi-population genetic algorithm (SCMPGA), the algorithm of ANN evolutionary modeling based on SCMPGA, the application of SCMPGA in power load forecast system modeling and so on. The main achievements are as fellow:At first this thesis summarizes the current achievements of genetic algorithm, researches the dynamic properties of multi-modal function optimization problem (MFO). The author quantitative analyzes the competition properties among niche in MFO, qualitative analyzes that niches cheat is the cause of premature convergence in MFO.Traditinal GA is difficult in solving MFO. The thesis analyzes the species evulution in nature and civilization & science advancement. SCMPGA is proposed, in which schema operates is introduced. The schemas are memorized , learned, united, excluded among populations in SCMPGA. The method is new way to solve premature convergence in MFO.ANN evolutionary modeling is MFO. The vality of ANN model is evaluated by its generalizationary ability. Based on SCMPGA, the author designes two improved algorithms of ANN evolutionary modeling. One is improved niche GA (INGA) which adoptes niche technology, another is improved parallel GA (IPGA) which adoptes parallel computing method. The simulation indicates that the two new method has the superiority of computation complicacy, model performance.The thesis gives the work flow of ANN automatic modeling in power load forecast system based on IPGA, which implementes automatic analyzing, evoluating, modifying, designing model. Its application demonstrates that this method is effective.The achievements of the thesis have great theoretic and realistic significance in researching the evolution mechanism of genetic algorithms and its prematue convergence problem, advancing ANN evolutionary modeling method.