【作者】 李广强；

【导师】 滕弘飞；

【作者基本信息】 大连理工大学 ， 机械设计及理论， 2003， 博士

【摘要】 本文研究基于计算智能和人机交互的设计理论、方法及其在工程系统布局方案设计中的应用，其工程背景是课题组承担的我国航天器舱的布局方案设计和为徐州工程机械集团开发的履带式起重机QUY150的相关布局设计问题。求解复杂工程布局问题的难点是存在计算复杂性的组合爆炸和工程实用化的复杂性。该课题在理论上，属带性能约束的三维布局问题，涉及航天器设计、机械工程、计算机、数学、力学和系统工程等学科，为交叉学科前沿课题的基础理论和应用基础研究，具有NP难度；在实践上，具有广泛的工程应用背景，如航天器舱、工程机械、潜艇、船舶、坦克、水下悬浮工程、海上钻井平台、高速列车、组合机床多轴箱、机器人等的布局方案设计问题。它的退化问题为不带性能约束的布局问题，如板材套排等。无论从理论上还是工程实践上，这都是一个长久以来人们关注的亟待解决而又未很好解决的重要问题。 复杂工程系统布局方案设计问题的性能指标和约束中，通常既有“定量的”又有“定性的”。目前解决这个问题主要有两条途径：a．数学优化模型及其算法求解+符号模型及其人工智能求解；b．数学优化模型及其算法求解+人机交互或人机结合。前者在实际应用过程中存在一定的困难，主要问题在于对该工程复杂问题，人工智能方法的知识库、推理机和自学习规则不容易建立。特别是例如航天器等高科技复杂系统，其相关数据和资料对外保密，因而建立知识库等将更为困难。此外，完全脱离人的参与将很难达到相关理论和方法的实用化。由此，结合国情，本文采用后者的求解方法，主要进行了以下的研究工作。 (1)综述了布局问题，尤其是复杂布局问题的类型、求解方法的研究现状和发展趋势，表明了研究优秀的计算智能类算法，同时加入人机结合或人机合作的思想，用以解决工程复杂布局方案设计问题的可行性和有效性。 (2)提出了并行混合蚂蚁免疫算法(PHAIA)，用以克服并行遗传算法(PGA)的早熟和收敛慢等两大缺陷。它对PGA主要进行了以下改进：a．利用混沌思想产生初始群体，并依交叉和变异概率值对子群体进行分类；b．建立了基于蚂蚁算法(ACO)的基因组合算子模型，将蚂蚁算法与遗传算法在基因层面上相结合，采用增加蚂蚁组合算子以及与Powell法相混合的思想来加快收敛速度：c．引入免疫功能具有双重作用，一是由本文给出的免疫选择操作可有效防止早熟，二是通过基于免疫记忆的子群体信息交换策略可加速收敛；d．算法中的交叉、变异和蚂蚁组合等算子均采用了自适应的思想。文中的算例表明，PHAIA是有效的，优于传统的并行遗传算法。 (3)提出了布局模式的相关理论。布局模式指待布物间的相对位置关系。它是一个相当重要、值得深入研究的问题。本文以航天器舱的布局方案设计为工程背景，重点研究了装填布局模式的表达、识别、构造和应用。定义了同构、非同构布局模式和待布物的布局等价关系等概念，给出了关系矩阵和模式矩阵及其变换，描述了布局模式控制区和非同构度，提出了同构和非同构布局模式的识别及构造方法。同时，本文还改进并重新给出了同构、非同构布局模式猜想(LPC)，且用数值计算的方法验证了它的合理性。最后讨论了布局模式和猜想的若干应用。这些工作为缓解给定规模的装填问题求解时存在的大连理工大学博士学位论文摘要组合爆炸，构造高效的求解算法提供了启发和借鉴。 (4)根据“人机结合”与“人机合作”的理念，在PHAIA的基础上，充分考虑设计者的知识和经验，利用可视化(viSC)技术，借鉴钱志勤等提出的人机交互的遗传算法(HCIGA)，给出了可视化交互式蚂蚁免疫算法(vIA认).它可适时地将人设计的方案作为贡献的个体不断地提供给PHAIA，同算法构造的个体一起形成进化群体，再经算法进化寻优，如此循环，直至到达结束准则。VIAIA能使人和机器的智能在算法层面上结合，有利于人机各自发挥其特长，为优质高效地解决复杂工程布局问题提供了方便。viA认有下列特点:a.充分利用可视化技术，可把计算过程中有意义的中间数据，随进化过程动态而正确地显示出来，从而更有助于设计者全面把握计算，科学、合理地参与交互;b.针对装填布局问题，将非同构度等概念引入VIAIA之中，为解决“人机合作”在工程布局设计中的“可操作性”问题提供了有益的尝试。根据非同构条形图给出的信息，设计者可适时添加与优秀个体同构或非同构的人工个体，以帮助算法增强局部搜索或跳出局部最优;。.在交互时体现了不同用户模型的个性化解释，力求对于不同水平的设计者，算法都尽可能地体现出较高的性能。 (5)为工程实用化，本文给出了一个复杂工程系统布局方案设计的问题求解策略。它按照从实物模型化到模型实物化的求解路线，采用数学优化和仿真的复合模型，在人机合作的思想指导下，综合利用了本文提出的相关理论、方法以及模糊综合评判等方案评价手段。基于该策略，本文开发了相应的软件系统，并用于求解国际通信卫星困TELSAI’-m舱和返回式人造卫星回收舱的布局方案设计，以及履带式起重机QUY150上车构件的布局设计问题，取得了令人满意的结果。其中返回更多还原

【Abstract】 This dissertation studies the relevant theory and design methods based on computational intelligence & human-computer interaction and their applications to layout design of engineering systems. The engineering backgrounds of this dissertation are layout design of the national spacecraft modules that our institute undertakes and the cooperative development work of crawler crane QUY150 with Xuzhou Construction Machinery Group Inc. The difficulties of solving complex engineering layout problems lie in computational complexity (Combinatorial explosion will occur.) and the complexity of engineering practice. This subject belongs to three-dimensional layout problems with behavioral constraints theoretically and concerns the knowledge of spacecraft design, mechanical engineering, computer science, mathematics, mechanics and system engineering. So it can be ascribed as the frontier fundamental and applied research of cross-discipline subject with NP complexity theoretically. In term of engineering practice, this subject has extensive applications, such as layout design of spacecraft module, engineering machinery, submarine and shipping, tank, under-water suspension engineering, platform of marine drilling well, bullet train, multiple spindle box of machine tool, robot. Their degenerate problems are the layout problems without behavioral constraint such as the nesting of steel sheet. Although it is concerned for a long time and of great importance and urgency in terms of theory and practice, this subject is far from being solved satisfactorily.As a rule, there are both the quantitative and the qualitative performance indexes and constraints in layout design problems of complex engineering systems. At present, the main methods for solving this kind of problems are as follows: (a) Mathematical optimization model and solving it by computer algorithms + Symbolic model and solving it by artificial intelligence. (6) Mathematical optimization model and solving it by computer algorithms + Human-computer cooperation or human-computer interaction. But it is rather difficult to put the former method into practice for complex engineering systems. The reason lies in that it is not easy to create knowledge base, inference engine and self-learning rule by AI methods under the circumstances. It is more difficult for the high-technology complex systems (e.g. spacecrafts), because the relevant data and documents are confidently and they can’t be exchanged outside the research and development circle. Furthermore, the relevant theory and methods are hard to be put into engineering practice without the fellowship of human. As above stated, according to the state of our nation, this dissertation adopts the latter method and mainly carries out the following research works.1. By survey of the various types of layout problems (especially complex layout problems), as well as the present status and trend of their solution methods, it makes clear that the method is feasible and effective for solving layout design problems of complex engineering systems, which adopts computational intelligence algorithms together with human-computer cooperation.2. To overcome the two main defects, i.e. premature convergence and slow convergence rate, of the traditional parallel genetic algorithm (PGA), a parallel hybrid ant immune algorithm (PHAIA) is proposed in this dissertation. PHAIA makes some improvements on PGA asfollows. (a) Chaos initialization is adopted and subpopulations are classified as several types according to the values of crossover and mutation probability. (b) The model of gene combination operator, called ant combination, based on ant colony optimization (ACO) is established and it introduces ACO into PGA in the layer of genes. Hybridized with Powell method and added with ant combination operator can improve local searching performance of the algorithm considerably, (c) Introducing immunity theory into parallel genetic algorithm has double functions. One is that immune selector proposed in this dissertation can prevent algorithm更多还原