Gene Expression Programming with Dual Strategies and Neighborhood Search for Symbolic Regression Problems
Hu Peng,Lin Li,Changrong Mei,Changshou Deng,Xuezhi Yue,Zhijian Wu
DOI: https://doi.org/10.1016/j.asoc.2023.110616
IF: 8.7
2023-01-01
Applied Soft Computing
Abstract:As an evolutionary algorithm for automatic generation of computer programs, Gene expression programming (GEP) has received a lot of attention because it can generate concise and highly accurate solutions. However, GEP also has some unsatisfactory aspects, such as too many control parameters and a lack of balance between exploration and exploitation. Therefore, a new GEP with dual strategies and neighborhood search (DNSGEP) is proposed in this paper for symbolic regression problems. As the core part, a new multi-parent crossover operator with neighborhood search is proposed to replace the original transposition and recombination operators in GEP, which greatly reduces the excessive control parameters in GEP. To assist the multi-parent crossover operator in selecting parents, selection weight is used to assess the quality of individuals in the population. The multi-parent crossover operator combines selection weight to learn more from the better parents. In order to satisfy the different needs of global search and local search of the algorithm, two different crossover strategies are extended on the basis of the multi-parent crossover operator. Furthermore, a population distribution entropy strategy to GEP is proposed for determining when the algorithm should choose which crossover strategy to use. DNSGEP is compared with GEP and three GEP variants on 14 symbolic regression problems and 6 even parity problems. Experimental results show that DNSGEP has the success rate and accuracy that are fully superior to GEP, GEPADF and OGEP, not inferior to SLGEP.