A mathematical model for heat exchanger network optimization was established based on the stage wise superstructure.By analyzing the serious nonlinearity of the model,it is shown that two key factors ,namely,the matching mode and the exchanger area,have a strong impact on calculation efficiency and accuracy.Moreover,the heat exchanger network optimization problem belongs to the mixed integer non-linear programming problem,so,it is easy to fall into the local solution in optimization process.In order to relieve this limitation in the synthesis of heat exchanger networks,a novel approach which combines initial multi-point method with Monte Carlo method was presented.Effectively jumping out of the local solution to obtain global optimal solution by using this approach was validated by numerical examples.