The Fusion Of GEP And MEP Together With The Novel Decoding Evaluation Technology

Genetic programming (GP) as a new evolutionary computation branch has become a hotspot in the study of artificial intelligence. The traditional GP encoding with tree structure, sothe evolution must operate complex syntax tree, which makes the algorithm difficult to realizeand affects the search efficiency. Therefore, on the basis of GP researchers have proposedlinear encoding and graph structure encoding of computer programs. Linear encoding of GPbecomes the main research direction. This paper mainly discussed and improved the geneexpression programming (GEP) and multi expression programming (MEP) algorithm.GEP encoding method is simple and can solve the complicated problem. But GP stillexist some bad situation in solving complicated problems, one is not get the optimal result,the second is too long evolution generation, the main reason is that the gene diversity islimited. The most obvious advantage of MEP is that a chromosome contains multipleexpressions, and its disadvantage is that the decoding process is too complicated In order tosolve the above problems, this paper improved the GEP and MEP algorithm based on theprevious research result, the main work is as follows:1）This article put forward an idea of sharing evaluation value for the same genefragments in a MEP chromosome. Based on the encoding gene to assess, thegenotype-phenotype mapping process is discarded. This improvement can accelerate thespeed of individual assessment.2）Compared with genes in a single chromosome, the same gene shares the completedevaluation results. Several experiments have proved the above improved MEP algorithmenhanced the function minning ability.3）Analysis of the strengths and weaknesses of the GEP and MEP evolutionaryprogramming, this paper proposed a novel genetic programming approach: Multi GeneExpression Programming (MGEP) which gets benefits from the GEP and MEP algorithmthought. The idea is realized by changing the GEP decoding method. MGEP contains multipleexpressions in a single chromosome.4）Analysis of the standard GEP decoding, we know it established expression tree withbreadth-first principle. This paper introduced the depth-first way to create expression tree.This thesis introduced these two decoding strategy into the MGEP algorithm. It proves intheory that the depth-first way could be higher efficiency to decode and evaluation. 5）Through some experiment to compare the evolutionary performance of the depth-firstMGEP, the breadth-first MGEP, the standard GEP and MEP. All kinds of algorithm need tomine the same function, mainly from the evolutionary generation, individual diversity andaccuracy to contrast the parameters result. The experimental results demonstrated that MGEPalgorithm has stronger ability to be automatic function found. The reason for this is thatMGEP population contains rich individual diversity.