Abstract: This research analyzed relationship between heat exchanger networks areas and total cost，and the features of non-linear and non-convex in the heat exchanger networks. These characteristics resulted in many local solutions in optimization process. The conventional optimization methods based on gradient information could be affected greatly by the initial point，and easily traped into local solution which interrupts the optimization process in the heat exchanger network. This work developed differential evolution algorithm for optimization of heat exchanger networks synthesis based on the Grossman stage-wise superstructure，aiming at solving the optimization obstacle in the heat exchanger network. Compared with the results derived from Newton method，this method showed robust in optimizing heat exchanger networks problems，and demonstrated minimum influence by the initial points. Consequently，this method indicated good stability and effectiveness，it could help in reducing total cost in productive process.

Key words: heat exchanger networks, initial point, differential evolution, convergence stability

摘要： 通过分析换热网络换热器面积与综合费用之间的关系，证明了连续变量引起的换热网络非线性及非凸等特性，这些特性造成换热网络局部最优解众多。而传统的基于梯度的优化算法受初始点影响很大，容易陷入局部最优解，是换热网络最优化的障碍所在。针对优化换热网络问题难以稳定收敛的角度出发，在换热网络分级超结构模型的基础上，研究了微分进化算法求解换热网络问题的鲁棒性。通过具体的算例，在固定结构下与经典的牛顿法比较，结果表明，该方法优化结果不受初始点影响，具有很好的鲁棒性，并且年综合费用得到进一步降低。用于工业生产过程中，可以稳定有效地节约成本。

关键词: 换热网络, 局部最优解, 微分进化算法, 收敛稳定性