Due to non-convex and non-continuous complexity of MINLP problems, heuristic methods are subject to numerous local optima. A novel random walking algorithm with compulsive evolution (RWCE) is an efficient global optimization method for HENS, where probability δ of accepting imperfect network is able to keep population diversity and jump out of local optima effectively during evolution process. In this paper, strong ability of parameter δ to detect promising structure is proved. Structure of evolution process is compared and analyzed under different parameters. Furthermore, a strategy for accepting imperfect network oriented with number of heat exchangers is proposed to improve computational time and evolution ability of optimization. The strategy was applied to several cases taken from literature. It shows better performance of global optimization.