For complex function mining, traditional gene expression programming (GEP) needs large number of evolutionary generations and may plunge into local optimum. To solve this problem, this paper presents a novel evolutionary algorithm based on multiple expression genes programming (MEGP). The main contributions include: (a) provide a novel gene hierarchical representation model to encode solutions of complex function finding; (b) propose a chromosome architecture that allows of a genome with multiple candidate expressions; (c) theoretically analyze and compare the expression space of MEGP algorithm with traditional GEP; (d) implement the MEGP algorithm and the chromosome fitness evaluation algorithm. Extensive experiments show that the success rate of MEPG is 2-4 times of traditional GEP.
