【作者】 徐歆尧； 徐德；

【Author】 Xinyao Xu;De Xu;Research Center of Precision Sensing and Control,Institute of Automation,Chinese Academy of Sciences;School of Artificial Intelligence,University of Chinese Academy of Sciences;

【机构】 中国科学院自动化研究所精密感知与控制研究中心； 中国科学院大学人工智能学院；

【摘要】 工业数据质量的好坏对数据分析与建模有很大影响。由于从工业现场采集的大量数据难免存在不同程度的缺失,采用合理的数据填补算法将有助于数据的建模与分析。实际工业过程通常包含多种生产工序,监测数据大多是非线性的多变量时间序列,且往往具有明显的动态特征。针对传统数据填补算法对工业过程动态特征建模能力不足的问题,本文采用sequence-to-sequence结构的门控循环单元(Gated Recurrent Unit,GRU)网络对过程数据进行建模,并在此基础上实现对多变量监测序列的缺失数据填补。文章基于某饮料无菌灌装生产线的温度监测数据验证了算法的有效性。更多还原

【Abstract】 The quality of collected industrial data affect the quality of data analysis and modeling. Data missing phenomenon exists generally in large amount of data collected from industrial processes. Using proper imputation algorithms will help data analysis and modeling. Since industrial production processes usually involve multiple operation modes, the monitoring data are often observed non-linear multi-variable time series, and often with dynamic characteristics. In order to solve the drawback of insufficient modeling ability of traditional imputation algorithms to model the dynamic characteristics of industrial processes, a sequence-to-sequence auto-encoder networks based on gated recurrent unit(GRU) network was constructed to model the process. Then, a missing value imputation method was proposed based on this model. Experiments based on the monitoring data collected from a beverage aseptic filling production line validated the effectiveness of the method.更多还原