MERGE: A Novel Evolutionary Algorithm Based on Multi Expression Gene Programming

Gene expression programming (GEP) is a new member in genetic computing. The traditional GEP lacks the power to handle very complex function mining problems due to its limited express capability. To solve the problem, this paper presents a new evolutionary algorithm named multi expression gene programming (MERGE). The main contributions include: (a) Provides a novel hierarchical gene encoding and decoding model; (b) Proposes a chromosome architecture that allows of a genome with multiple candidate expressions; (c) Implements MERGE algorithm and gene fitness evaluation algorithm. (d) Gives extensive experiments to show that MERGE outperforms the traditional GEP. Furthermore, When mining complex functions, the success rate of MERGE is 3-5 times of GEP, the average number of generation of successful evolution is 87% higher than GEP, and the average minimum generation of successful evolution of MERGE is reduced to 0.4% of GEP.