Abstract: To solve the problem of low data quality and availability caused by abnormal data during data acquisition of loom equipment,a method of processing loom abnormal data based on probability distribution and XGBoost decision algorithm was proposed.The time series variation difference between adjacent data points of each parameter of the loom were obtained,and the confidence interval reflecting the approximate variation trend of each parameter was got.Based on the degree of correlation between loom parameters,Bayesian network was used to infer the probability distribution of normal data points,deviating abnormal points and inactive abnormal points of loom,and the time point and type of abnormal data were located.Based on XGBoost decision algorithm,the missing bits caused by abnormal data from the loom were repaired.Taking the data collected from the loom equipment of a textile enterprise in Shijiazhuang as an example,the result showed that the average abnormal recognition rate of the proposed loom abnormal data recognition method was increased by 13.49% and 19.09% respectively compared with the decision tree and K-Nearest Neighbor.The proposed model could improve the repair accuracy.The fitting coefficients of weaving output,beating times,running efficiency and running speed were 0.964 6,0.956 3,0.983 2 and 0.973 6 respectively.

Key words: air-jet loom, abnormal data, probability distribution, XGBoost, data recovery

摘要： 为了解决对织机设备进行数据采集时出现异常数据导致数据质量低下、可用率低的问题,提出一种基于概率分布和XGBoost决策算法的织机异常数据处理方法。首先,获取织机各参数相邻数据点的时序变化差值,得到反映各参数近似变化趋势的可信区间;其次,基于织机参数间的相关性程度,利用贝叶斯网络对织机正常数据点、偏离异常点和不活跃异常点3类数据类型的概率分布进行推理,定位异常数据发生的时刻点和异常类型;再次,基于XGBoost决策算法对织机异常数据造成的数据缺失位进行修复。最后,以某纺织企业织机设备的采集数据为例进行分析表明,与决策树和K最近邻算法相比,所提织机异常数据识别方法的平均异常识别率分别提升13.80%,19.09%;所提织机缺失数据修复模型能提升修复准确度,对织机织布产量、打纬次数、运行效率和运行车速的拟合系数分别为0.9646,0.9563,0.9832,0.9736。

关键词: 喷气织机, 异常数据, 概率分布, XGBoost, 数据修复