Application And Research Of Gene Expression Programming In Function Mining

This paper first reviewed development, application background, basic characteristic of evolution computation and function mining; summarized each essential technique of the classical GEP which mainly included individual composition, coding rule, K-expression, fitness function, selection operator, mutation operator, IS operator, recombination operator, multi-genes chromosome, constants and so on. This paper also in detail elaborated gene expression programming characteristic; analyzed the classical GEP insufficient in the maintenance population diversity and the protection optimal solution and the reason of prematurity. In order to overcome these shortcoming, this paper put forward an improved GEP algorithm:GEP-FM which adopted head, body, tail structure and self-adaptive mutation operator; also put forward new fitness function and VLCF algorithm which computes fitness based on the gene valid length; the theoretical analysis including complexity analysis and convergence analysis of algorithm was also carried on. In order to verify the accuracy and the validity about GEP-FM algorithm, this paper applied GEP-FM to function mining. Through mining one variable function, two or three variables function, several variables complex function, those experiment results indicated GEP-FM has good performance and is superior to traditional algorithm and classical GEP, the models set up by GEP-FM are better than the models set up by traditional algorithm and classical GEP, and its prediction accuracy is high.