A Study on Neighborhood and Temperature in Multi-step Crossover Fusion for Tree Structure

Abstract

Multi-step crossover fusion (MSXF) is one of promising crossover methods for solving combinatorial optimization problems. MSXF performs multi-step local search from a parent in the direction approaching the other parent. In the transaction process of the local search, a neighborhood solution is stochastically accepted as the next solution, according to Metropolis criterion. To improve the search performance of MSXF, a sophisticated neighborhood structure and an appropriate scheduling of temperature parameter are required. In our previous work, we proposed an improved neighborhood generation method that promotes the recombination of parents' preferable characteristics for tree structures. In this paper, we validate the efficacy of the proposed method and evaluate the temperature scheduling in several symbolic regression problem instances.