Advanced Textile Technology

   

Abnormal data processing of spinning machines based on ARIMA-Bayesian network and hybrid repair method

  

  • Published:2025-03-19

基于ARIMA-贝叶斯网络与混合修复方法的纺纱机异常数据处理

  

Abstract: "In response to the issue of abnormal data arising from communication failures, data acquisition and transmission anomalies in the data collection process of spinning machines, this paper proposes a dual-dimensional abnormal data identification and hybrid missing data repair method to enhance data completeness and accuracy. Therefore, how to effectively identify abnormal data and accurately repair missing data is a key issue to ensure the normal operation of textile equipment and improve the precision of data analysis. The dual-dimensional abnormal data identification method proposed in this paper combines analytical models from both the time dimension and the parameter dimension, fully leveraging the complementary advantages of the autoregressive integrated moving average (ARIMA) model and the Bayesian network model. In the time dimension, the ARIMA model is used to analyze the single-parameter time series data to identify anomalies in individual parameters over time. In the parameter dimension, this paper uses the Bayesian network model to construct causal relationships among multiple parameters, thereby detecting anomalies related to the interactions between different parameters. Since there is usually a strong correlation between multiple parameters of the spinning machine, it may not be possible to fully identify anomalies by analyzing a parameter alone. This paper calculates the Spearman rank correlation coefficient between parameters to construct a correlation matrix and generates a causal network graph based on this matrix, identifying anomalies at a specific time point that are inconsistent with changes in other parameters. After the abnormal data identification is completed, in order to ensure data completeness and accuracy, this paper proposes a hybrid repair method to fill in missing data. According to the characteristics of the spinning machine data, the K-nearest neighbors (KNN) algorithm is used to repair stable fluctuating data. The KNN algorithm fills in missing values by comparing the similarity between the missing point and its neighboring points and selecting the closest neighboring value. For continuously increasing data, a piecewise linear regression combined with a sliding window prediction method is used to repair. The experimental results show that the dual-dimensional abnormal data identification method proposed in this paper achieves an identification rate of 97.58 %, and the average fitting coefficient of the hybrid repair method reaches 0.9614, which verifies the effectiveness and reliability of the proposed method. The method presented in this paper can identify abnormal data more comprehensively and repair different types of missing data accurately, providing significant technical support for the intelligent production of spinning machines and the optimization of spinning processes."

Key words: spinning machine, abnormal data identification, data repair, ARIMA model, Bayesian network model, time series data

摘要: 纺纱机在数据采集中因通信故障、数据采集与传输错误常引发数据异常问题,对此提出了一种双维度异常数据识别与缺失数据混合修复的方法。首先,根据不同模型互补优势,从时间维度和参数维度分别对异常数据进行分析,利用自回归移动平均模型对单参数时间序列进行异常数据识别,分析单个参数在时间维度上的变化趋势;其次,利用贝叶斯网络模型对多参数时间序列进行异常数据识别,建模不同参数之间的关联关系,识别时间点上的关联异常;最后,根据数据特性,利用K近邻算法修复稳定波动型数据,采用分段线性回归结合滑动窗口的混合预测方法修复连续增长性型数据,确保数据完整性与准确性。结果表明:双维度异常数据识别方法识别率可达97.58%,混合修复方法拟合系数平均达到了0.9614,有效提升异常数据处理的精度。研究结果可为纺纱机数据分析及后续优化提供可靠的数据支持。

关键词: 纺纱机, 异常数据识别, 数据修复, ARIMA模型, 贝叶斯网络模型, 时间序列数据

CLC Number: