离散型生产大数据环境下的异常事件回溯模型研究

When planning "made in China 2025", the Chinese Academy of Engineering puts forward the development strategy of discrete intelligent factory. The key of intelligent factory lies in intelligence, and the core of intelligence lies in processing and application of Big Data. Because the traditional relational data paradigm is difficult to support the processing of massive multi-source heterogeneous production information, it is difficult to trace abnormal events in discrete production under Big Data environment. In order to solve this problem, a new Big Data paradigm for discrete production is proposed in this project. This project are proposed by using deep learning algorithm (DBN) to mine the organic aggregation rules of production information and researching on classification and aggregation model of massive multi-source heterogeneous discrete production information and the reduction of redundant information. In the framework of this paradigm, a new model for production anomaly event backtracking is established by studying and improving variable neighborhood ant colony and biological immune algorithm. Combined with the new paradigm, based on the study of the model strategy, the methods of abnormal event backtracking optimization and model self-evolution are proposed. Finally, a prototype of abnormal event backtracking model in discrete production under Big Data environment is developed and verified by engineering. Through the implementation of this project, the following two points can be realized in the discrete production environment of Big Data: 1) integration of massive multi-source heterogeneous data, 2) global and efficient backtracking of production exception events.

中国工程院在规划《中国制造2025》时，提出离散型智能工厂发展战略。智能工厂的关键是智能化，而智能化的核心之一就是大数据的处理和应用。本项目针对离散型生产大数据环境下，传统关系型数据范式难以支撑海量多源异构信息处理的难题，导致异常事件回溯困难的问题，采用深度学习算法挖掘生产信息的有机聚合规则，研究海量多源异构生产信息的分类方式、聚合模式以及冗余信息消减方法，提出一种离散型生产大数据的新范式；在新范式的框架下，研究生产异常事件回溯建模策略，并通过改进变邻域蚁群和生物免疫算法，提出异常事件回溯方法优化和模型的自进化，以构建异常事件回溯新模型。最后，研制离散型生产大数据环境下的异常事件回溯模型原型并工程验证。通过实施本项目，可实现离散型生产大数据环境下：（1）海量多源异构数据的集成整合，（2）生产异常事件的全局高效回溯。

本项目针对离散型生产大数据环境下，传统关系型数据范式难以支撑海量多源异构信息处理的难题，导致异常事件回溯困难的问题，研究了：1)非结构化数据的故障图谱创建，2)结构化数据的故障特征提取与分类，3)边云协同的数据采集与异常检测，4)多源数据聚合范式---M-source N-level，5)知识图谱构建及物流设备路径规划，6)异常事件回溯建模策略（路径规划算法优化）实验。研究建立知识和数据的故障图谱，有利于对本领域的经验知识进行组织和管理，同时也能辅助故障知识的传承与积累，使故障知识及时持久化和数字化，实现隐性知识显性表达、显性知识立体表达，有利于打通故障知识边界。融合的知识和数据能从冗余的数据以及知识经验文本中抽取有价值的信息，从而表达数据与知识之间的内在关联关系，再通过关联性进行故障和异常的诊断与回溯分析，为智能化故障诊断提供知识导流和推荐支持。研究内容1）、2）、4）提出的方法和范式对于故障诊断领域具有理论创新意义；在此基础上，研究内容3）、5）、6）提出的方法和优化实验，具有技术创新意义和工程应用价值。本项目的研究结果，从微观角度看，可以完成生产车间异常事件回溯；从宏观角度来说，有助于离散型生产车间的智能改造与升级，特别是为数字孪生车间的构建提供了支撑模型。