【作者】 刘洪谦；

【导师】 麻德贤；

【作者基本信息】 北京化工大学 ， 化学工程， 1999， 博士

【摘要】 过程系统综合问题大都属于NLP／MINLP范畴。工程过程中的非线性约束、经验性和逻辑性约束使这些问题很难全局求解。 针对上述情况，本论文主要作了以下方面的工作： 1) 用信息提取技术对遗传算法进行了改进，增强了遗传算法的全局遍历能力，提高了遗传算法对多峰、奇异和不连续函数的求解能力。根据遗传算法的基本原理，指出：在遗传算法中，种群以及种群中的样本应是搜索空间的信息代表点：遗传操作应是随机和整体的统一体。利用杂交系数矩阵和变异系数矩阵对种群进行整体杂交和整体变异，用并行自适应策略增强算法的全局遍历能力，有效地克服”遗传漂移”造成的“早熟”现象。 对十八个多峰、奇异和不连续测试函数的有效求解，说明：在小样本、小并行空间条件下，信息提取和改进遗传算法能有效地实现问题的全局优化。 对有再循环系统的三级过程设计、反应网络综合、已知网络结构的换热网络设计、多单元过程设计等过程系统综合问题的全局优化，显示了信息提取和改进遗传算法对NLPMINLP问题全局优化的能力。计算结果表明：无论是计算效率、计算结果还是计算精度，信息提取和改进遗传算法要比其他优化方法更加有效。 2) 对分离序列最优综合策略进行了研究，提出了一种分离序列超结构的简洁表示方法和相邻分离序列的产生规则。对五组份和十二组份分离序列的最优综合，显示了信息提取和改进遗传算法对该类NP难问题的求解能力。 3) 根据换热网络的实质指出：换热网络综合是一个二分图指派问题，其实质是在冷热物流间构造多个开环或闭环的Hamiton圈，即是一个在冷热物流间寻找二分图最短匹配路径的过程。 根据以上分析，提出了换热网络基本回路的识别策略。通过基本回路的识别，能有效地识别网络中的全部回路。 对实际换热网络的有效识别，验证了以上识别策略的有效性。 4) 本论文在“转运模型”的基础上，根据二分图的基本原理，将匹配负荷当作自变量，将公用工程消耗当作网络能量回收的松弛变量，利用信息提取和改更多还原

【Abstract】 Process system synthesis and analysis are one of the most important parts of process system engineering (PSE). Constraints, which are non-linear or from logical strategy and experience, make it difficult to obtain the global optimization of process synthesis and analysis.Because of the complexity of nonlinear programming (NLP) / Mixed Integer Nonlinear Programming (MINLP), and the lack of effective mathematics algorithms for large scale NLP/MINLP, the optimization algorithms to process synthesis are still considered the most difficult.The main contents of this dissertation are below:1) Information extraction technique has been introduced to improve genetic algorithms. It was pointed out that population and the individuals of a population are information points, which represents the useful information about the available space. With parameter matrix of mutation and parameter matrix of crossover, generalizing from information extraction and population could be mutated and crossovered on the whole. Genetic algorithms could be considered as a combination operation, which is stochastic and entire.Parallel and adaptive strategy enhanced the ergodic ability of the algorithm to available space, and crossover overcame effectively the information lose resulted from "Genetic Drift".The solutions to multiple-peak and rigid functions, or to NLP / MINLP problems such as process design and synthesize of reactor network, show us the efficiency of this improvement. Especially, to some problems, the results are rather more satisfied.2) The synthesis of heat exchanger networks (HENs) is a problem of a bipartite graph, whose heart to constructing open and close Hamilton circles between cold and hot streams. On the other hand, it could be considered as searching the shortest bipartite match loop among cold and hot streams.In this dissertation, a loop identification strategy to base loop in HENs is put更多还原