基于模式本体和机器学习的信息物理系统建模与优化

发布时间：2018-03-04 12:28

本文选题：信息物理系统　切入点：建模与优化　出处：《华东师范大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,信息物理系统受到了学术界和工业界越来越多的关注,其复杂度使其建模与优化问题成为了一个研究热点。信息物理系统自身具有的混成性、不确定性等特点,给建模工作带来了一定的困难。而且,信息物理系统的领域相关性,使得其建模的质量严重依赖于设计人员的经验。另外,信息物理系统所处环境中可控和不可控因素并存,如何针对这些因素进行系统优化也困扰着研究人员。针对这些问题,本文提出了基于模式本体和机器学习的信息物理系统建模和优化方法,主要工作包括:·提出了一种基于随机混成自动机和模式本体的信息物理系统建模方法。该方法使用随机混成自动机网络对信息物理系统进行建模,抽取随机混成自动机的基本概念,构建上层模式本体。对上层模式本体进行实例化,构建领域模式本体。给出了基于模式本体的建模方法,并在此基础上使用统计模型检测工具UPPAAL-SMC对系统性能表现进行量化评估。·针对信息物理系统所处物理世界中的不可控因素,提出了一种基于支持向量分类的资源调度策略选择方法。该方法以多种用户行为下各种调度策略的性能表现评估结果为原始数据,经过数据预处理后,进行分类模型的训练,建立分类模型,并给出了应用分类模型进行资源调度策略选择的方法。·针对信息物理系统物理世界中的可控因素,提出了一种基于支持向量回归和帕累托优化的参数配置多目标优化方法。该方法首先选取待优化参数,进行参数实例的生成。选取部分实例的量化评估结果作为训练集,建立参数实例与系统性能表现之间的关系模型,对预测集中参数实例的性能表现进行预测,从而快速得到大量参数配置在多个目标下的性能表现,最终基于帕累托优化方法得到多目标下系统的最优化参数配置。最后,本文使用一个典型的信息物理系统——智能建筑系统对上述方法进行了详细的实例验证。实验结果表明,本文提出的基于模式本体的信息物理系统建模方法可以重用领域知识,帮助系统设计人员高效建立较高质量的系统模型;基于机器学习的信息物理系统优化方法,可以帮助系统设计人员快速得到相对优化的系统设计方案,满足相应的非功能需求。
[Abstract]:In recent years, more and more attention has been paid to information physics systems in academia and industry. The complexity of information physics systems makes modeling and optimization problems become a research hotspot. Information physics systems have their own characteristics such as mixing, uncertainty and so on. It brings some difficulties to the modeling work. Moreover, the domain relativity of the information physics system makes the modeling quality depend heavily on the designer's experience. In addition, the controllable and uncontrollable factors coexist in the environment of the information physics system. How to optimize the system based on these factors also puzzles the researchers. Aiming at these problems, a method of modeling and optimizing information physics system based on pattern ontology and machine learning is proposed in this paper. The main work includes: 路A modeling method of information physics system based on random hybrid automata and pattern ontology is proposed, which uses random hybrid automaton network to model information physics system. The basic concept of random hybrid automata is extracted, the upper pattern ontology is constructed, the upper pattern ontology is instantiated, the domain pattern ontology is constructed, and the modeling method based on pattern ontology is presented. On this basis, the statistical model detection tool UPPAAL-SMC is used to quantitatively evaluate the performance of the system. 路aiming at the uncontrollable factors in the physical world of the information physics system, This paper presents a resource scheduling policy selection method based on support vector classification, which uses the performance evaluation results of various scheduling strategies under various user behaviors as raw data. After data preprocessing, the classification model is trained. The classification model is established, and the method of resource scheduling policy selection by applying the classification model is given. A multi-objective optimization method for parameter configuration based on support vector regression and Pareto optimization is proposed. The relationship model between parameter instances and system performance is established to predict the performance of parameter instances in the prediction set, and the performance performance of a large number of parameter configurations under multiple targets is obtained quickly. Finally, based on the Pareto optimization method, the optimal parameter configuration of the multi-objective system is obtained. Finally, a typical information physical system-Intelligent Building system is used to verify the above method in detail. The experimental results show that, The information physics system modeling method based on pattern ontology can reuse domain knowledge and help system designers to build high quality system model efficiently. It can help the system designers to quickly obtain the relative optimization of the system design scheme to meet the corresponding non-functional requirements.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP181;TP29

【参考文献】