Random walk algorithm with compulsive evolution(RWCE) is an effective method for the optimization of heat exchanger network with advantage of simple evolutionary strategy and less control parameters. The probability of accepting improper solution has important influence on individual jumping out of local optimal solution. We count times of total annual cost decrease in a certain iteration interval in the later stage of optimization, and analyze influence of probability of accepting improper solution in the optimization process. A heat exchange network optimization method is proposed to adjust probability of accepting improper solution with difference between structure of improper solution and original structure. It strengthens evolutionary ability of the algorithm. Effectiveness of the strategy is shown with several examples.