【作者】 赵楠；

【导师】 郭伟；

【作者基本信息】 天津大学 ， 机械制造及其自动化， 2010， 博士

【摘要】 面对制造行业快速发展、企业兼并与重组不断的局面,如何从供应链存在的管理问题出发,在对供应链管理水平进行分析评价和动态监控的基础上有针对性的实施改善,进而打造出敏捷、精益、绿色的供应链已成为国内诸多大企业走向国际化的管理阻碍。本文针对制造业复杂供应链绩效问题,依照知识获取的路径,提出一套基于机器学习的智能分析方法。主要讨论供应链绩效数据质量控制、供应链绩效关键因素提取以及供应链绩效管理问题的诊断推理,以实现供应链绩效分析的自动化与智能化。通过研究,本文获得以下一些研究成果。(1)提出了供应链绩效数据质量控制方法。首先,基于全面数据质量管理的思想,结合抽样调查理论,给出一种多阶段闭环的数据质量控制流程。重点探讨了样本数据中的缺失值插补问题,基于缺失数据处理技术、矩阵分解理论以及数据挖掘技术,给出两种有效的插补算法:基于非负矩阵分解的插补算法和基于分类回归树的插补算法,通过数值实验证明了算法的有效性。(2)提出一种基于选择性集成的稳健主成分算法,实现供应链关键因素的提取。针对供应链绩效分析的复杂性以及调查数据的不确定性,基于稳健统计理论,实现了投影寻踪主成分算法的稳健化。基于Bagging集成学习算法和选择性集成思想,利用信息融合技术,实现了主成分分析的选择性集成,从而拓展了集成学习在无监督算法中的应用。通过数值实验和实际供应链绩效数据分析证明了该方法的有效性。(3)应用贝叶斯网实现供应链绩效问题诊断分析。首先分别给出基于模糊解释结构模型和Vague集合的贝叶斯网构建方法,有利于利用专家的先验知识构建模型。通过贝叶斯网的学习算法,得到供应链绩效关键因素对管理问题的影响路径。通过实际数据的分析以及与粗糙集等多种算法对比证明了算法的有效性。(4)开展了针对国内典型制造企业供应链绩效的智能分析工作。以本文提出的相关理论方法为依托,对国内大型发动机制造企业C公司所在供应链进行绩效智能分析,包括:供应链绩效数据清洗、关键因素提取、影响路径获取。借鉴管理咨询的手段,对绩效进行全面分析。通过对比两种方法的分析结论证明本文所提出的理论方法具备可行性和有效性,可以实现供应链绩效分析的自动化、智能化。最后,基于本文提出的部分理论方法,开发了供应链绩效智能分析系统。更多还原

【Abstract】 Considering the fast development of manufacturing, enterprise merging and recombination, creating the agile, lean, green supply chain, which is based on analysis and dynamic monitoring, is the main obstacle of the internationalization for domestic large enterprises. A set of intelligent analysis method based on machine learning theory was proposed, according to the manufacturing complex supply chain performance problems and the path of KDD. In order to realize supply chain performance analysis automation and intelligence, the supply chain performance data quality control, the key factors of supply chain performance and supply chain performance management problem in extracting diagnosis reasoning are discussed in this study. The main research results are as following.(1) The supply chain performance data quality control was proposed. First, based on total data quality management theory, combining the sampling theory, a multi-stage closed-loop data quality control process has been given. Based on the missing data processing technology, decomposing matrix theory, and data mining technology, two kinds of effective imputation algorithm has been given, which based on the decomposition of the nonnegative matrices factorization and CART. The effectiveness of the algorithm is demonstrated by the numerical experiments.(2) An algorithm of supply chain key factors extraction based on selective ensemble of robust principal component was proposed. According to the complexity of the supply chain performance analysis and investigation data, based on the robust statistic theory, realize the projection pursuit of main component algorithm is stable. Based on Bagging selective ensemble learning algorithm, information fusion technology, realize the principal component analysis selective ensemble.(3) Application Bayesian network to realize supply chain performance diagnosis. Bayesian network construction methods based on fuzzy ISM and Vague set have been given. Through the Bayesian network learning algorithm, the supply chain performance of management problems of key factors affect path. The validity of the algorithm is verified through the analysis of the actual data and rough sets of comparison algorithm.(4) Typical domestic manufacturing supply chain performance modeling and analysis practice. Based on the theory and method propose in this paper, a supply chain performance intelligent is analyzed for a domestic engine manufacturing enterprise. By comparing the two kinds of methods of analysis results, the validity and feasibility of supply chain performance analysis is verified, and the automation and intelligence of supply chain performance can be realized. Based on parts of method proposed in this paper, the supply chain performance intelligence analysis system has been developed.更多还原