Abstract: To improve the practical feasibility,computational efficiency and real-time response ability of flexible job shop dynamic scheduling,a method of mining dispatching rules from scheduling-related historical data with the characteristics of big industrial data was proposed.The data preprocessing of data collection,cleaning,integration and screening was completed by open source big data technology.Considering that the change of disturbance environment would change the influence of production attributes on scheduling decisions,the data sets relating to previous schedules were clustered based on disturbance attributes.In the mining of dispatching rules,an improved random forest algorithm was proposed.Case studies proved the feasibility and effectiveness of the proposed method.

Key words: flexible job shop, dynamic scheduling, dispatching rule mining, clustering, improved random forest algorithm, industrial big data

摘要： 为提高柔性作业车间动态调度的实际可操作性、计算效率与对车间扰动的实时响应能力,提出一种从具有工业大数据特点的调度相关历史数据中挖掘调度规则的方法。该方法在进行数据预处理时,结合开源大数据技术完成数据采集、清洗、整合与筛选。同时,考虑到动态调度问题中,扰动环境变化会改变生产属性对于调度决策的影响程度,对调度相关历史数据集合进行基于扰动属性的聚类,从而对不同扰动环境下做出的调度决策所产生的数据集合进行合理划分。此外,在调度规则挖掘中,提出了改进的随机森林算法。通过实例研究证明了该调度规则挖掘方法的可行性与有效性。

关键词: 柔性作业车间, 动态调度, 调度规则挖掘, 聚类, 改进随机森林算法, 工业大数据