摘 要：启发式算法在处理换热网络问题时具有可操作性强、搜索域大等优点，但由于局部解众多，算法很难寻得全局最优。本研究基于强制进化随机游走算法，以费用下降为强制进化方向，按照换热量最小，公用工程、流股匹配回路是否存在的优先顺序确定摄动对象；并以一定的概率对其进行随机地换热量线性变化或直接消去。重复寻找原结构下的更优分布或者新的网络结构。此外，引入梯度近似公式提高随机摄动方法的搜索精度。最后，通过计算10股流和20股流算例得到相较文献更低的年综合费用，分别为5 586 942和1 739 079 $/a，证明该方法能够有效地促进换热网络结构进化，得到更优的网络结构。

Abstract:The heuristic algorithm features such merits as a strong operability and large searching domain when dealing with the network problems, however, due to numerous local solutions, it is very difficult for the algorithm in question to obtain an overall optimum through searching. Based on the compulsively evolved random walk algorithm, with a decline in costs serving as the compulsive evolution direction, the perturbation objects were determined according to the priority sequence of a smallest amount of heat exchanged, public engineering projects and whether or not a circuit to match any strand of a stream was present and in the meantime, at a given probability, a linear transformation of the amount of heat exchanged was performed randomly or such amount of heat directly cancelled. A more optimal distribution or a novel structure of the network was repeatedly searched in the original structure. A better distribution or a new network structure in the original structure was repeatedly searched. In addition, an approximate gradient calculation formula was introduced to enhance the searching precision of the stochastic perturbation method. Finally, through using the calculation cases involving ten strands of a stream and twenty strands of a stream, much lower annual comprehensive costs than those in the literatures were obtained, being 5,586,942 and 1,739,079 $/a respectively, indicating that the method in question can promote the evolution of the structure of the heat exchange network and obtain an even better structure of the network.

TK124

国家自然科学基金（51176125）；上海市地方能力院校建设项目 (16060502600)