基于数据驱动的离散制造系统智能化调度建模与决策方法研究

With the development of the automation and informatization of Chinese manufacturing enterprises, huge amount of historical data has been generated and stored. The implied information includes real situation of shop floors and valuable scheduling knowledge. How to obtain useful information from the data through data mining technique, and thus to drive decision-making is the key to intelligent scheduling. This project focuses on Job shop scheduling problem. Various data analysis models will be developed through machine learning techniques in order to figure out the principle properties and to track the real status of the system. The objective is to realize the “self awareness” of the system. Based on the data analysis models, the scheduling model will be built to formulate the key decisions, as well as the relation between scheduling decisions and system status. An intelligent algorithm with “self-learning” function will be developed through the analysis and usage of historical data, simulation data, as well as scheduling experience and knowledge. Finally, the methodology is validated through real case study. .Through the project research, a comprehensive methodology will be developed to support the intelligent scheduling of manufacturing systems. The research results of this project have an important theoretical and realistic contribution in improving the decision-making and the production capabilities of Chinese manufacturing enterprises.

随着制造企业自动化及信息化程度的提高，产生并存储了大量历史数据，这些数据隐含了反映实际调度环境特点及调度知识的大量有效信息。如何从这些数据中挖掘信息，并驱动决策成为实现智能化调度的关键。本项目以Job shop型离散制造系统为对象，基于机器学习技术建立多种数据分析模型，从大量制造数据中揭示系统的运行规律，并真实反映系统的状态变化，实现系统的“自我认知”。在此基础上研究融合数据分析模型的生产调度建模方法，以准确定义关键决策问题，以及调度决策与系统状态变化之间的逻辑关系。通过分析和利用生产历史数据、调度决策仿真数据，以及相关经验和知识，设计开发具有“自我学习”功能的智能化调度算法，实现调度知识的传承和利用。通过实证研究对所提出的方法进行进一步验证，最终系统而完整地提出一套针对典型离散制造系统进行智能化调度决策的理论和方法，对于提升企业的科学决策水平及制造能力具有重要的学术价值和现实意义。

本项目针对离散制造系统智能化调度中的基本决策问题，基于隐式半马尔可夫模型建立设备状态识别模型，实现设备状态识别及加工工时预测。在此基础上，研究考虑设备状态的离散制造系统智能化调度方法，分别研究具有时变效应和质量损失的制造系统决策方法，考虑具有离散和连续设备状态变化特征的两类典型生产环境，实现生产及维护联合优化。针对设备状态具有连续性变化特点的制造系统，使用可靠性定义设备状态变化规律，建立不同生产场景中为实现不同优化目标的生产调度模型。基于问题特性分析，提出融入机器学习方法的智能化优化算法求解大规模调度问题。针对设备状态呈现离散变化特征的制造系统，提出建立在设备状态识别结果与生产调度关键参数（如加工工时）关联关系基础上的动态优化决策过程。最终系统而完整地提出一套针对离散制造系统进行智能化调度决策的理论和方法，促进离散制造系统运维和调度决策水平的提高。