基于分段聚合和卡尔曼滤波的纱线直径时间序列预测
来源:用户上传
作者:王延蒙 秦鹏 张文国
摘 要:为准确预测纱线直径,提高纱线质量预测的准确度,首先对纱线直径数据采样原理进行分析,对纱线样本片段分段聚合,利用聚合后的纱线直径值建立时间序列模型状态方程,采用自回归滑动平均模型ARMA(p,q)进行纱线直径和变异系数预测,然后利用卡尔曼滤波对预测值进行优化。通过实验对预测模型进行准确性验证,结果表明:卡尔曼滤波优化后预测的纱线直径均方根误差为2.68%,平均绝对百分比误差为6.71%;比对其他预测方法预测的条干不匀率,显示出良好的预测精度;模型泛化验证所选取的8个实验样本的检测结果均在乌斯特50%统计值内,同时纱线平均直径与理论直径之间的误差小于3%。这表明该预测模型对于在线预测纱线质量具有一定的准确性,为预测纱线质量提供一种新方法。
关键词:分段聚合;时间序列;卡尔曼滤波;纱线直径;数据预测
中图分类号:TS112.2
文献标志码:A
文章编号:1009-265X(2022)02-0041-07
收稿日期:20210303 网络出版日期:20210804
基金项目:山东省高等学校青年创新团队人才引育计划项目(2019189)
作者简介:王延蒙(1991-),男,山东菏泽人,讲师,硕士,主要从事纺织机械自动化设计方面的研究。
Yarn diameter time series prediction based on piecewisepolymerization and Kalman filter
WANG Yanmeng, QIN Peng, ZHANG Wenguo
(a.Department of Mechanical and Electrical Engineering; b.Jining MechanicalSystem Intelligent Research Institute, Jining Polytechnic, Jining 272037, China)
Abstract: For a more accurate yarn diameter prediction and accurate yarn quality prediction, the principle of yarn diameter data sampling was firstly analyzed, piecewise polymerization of yarn sample fragments was performed, and a time series model state equation was established based on the yarn diameter value after polymerization. Next, the yarn diameter and the coefficient of variation were predicted using autoregressive moving average model. Then the predicted value was optimized using Kalman filter. The accuracy of the prediction model was verified through experiments, and the results showed that the root mean square error of the yarn diameter predicted after Kalman filter optimization was 2.68%, with an average absolute percentage error of 6.71%. Compared with the yarn unevenness predicted by other methods, this method exhibited excellent prediction accuracy. The test results of the eight experimental samples selected for model generalization verification were all within 50% of Uster statistical value, with the error between the average yarn diameter and the theoretical diameter of less than 3%, indicating that the prediction model is accurate when applied to yarn quality online prediction. This prediction model can be used as a new method for yarn quality prediction.
Key words: piecewise polymerization; time series; Kalman filter; yarn diameter; data prediction
影纱线质量的主要因素为原棉质量和纺纱加工系统的工艺[1-2]。质量预测是控制纱线质量的常用方法手段,纱线质量预测可以起到降低成本、提高生产效率的作用。纱线直径是纱线质量指标的重要参数。目前,纱线质量预测常采用神经网络模型:Selvanayaki等[3]采用支持向量机(Support vector machines)的方法,将纱线强力的预测转化为凸二次规划问题;Mokhtar等[4]确定了纱线质量与影响因素的非线性关系模型,张羽彤等[5]、查刘根等[6]、李惠军等[7]优化设计BP神经网络预测条干不匀率;邢鹏程[8]改进了Apriori算法利用棉纤维各项指标预测纱线质量,杨建国等[9]使用粒子群算法和ELM算法相结合预测纱线的质量,袁利华[10]从HVI检测系统中选取棉纤维特性指标,利用RBF神经网络进行纱线质量预测,该方法选取的工艺参数多达14种,模型较为复杂。以上预测方法均为通过间接分析原棉参数和纺纱工艺性能之间的关系进行间接预测。部分学者利用时间序列模型分析了纱线直径的时间序列规律。Mohamed等[11]对棉纤维混合属性采用回归分析进行预测,袁汝旺[12]研究了纱线直径样本片段的相关特性,程立超[13]基于直接测量的纱线直径值建立时间序列模型,预测纱线直径,该方法忽略了纱线直径的随机性。时间序列模型不必分析原棉参数和工艺参数的影响,利用样本历史值预测新的样本值,但是时间序列预测具有短时性,且只能预测线性变化部分,预测精度较低。因此依据线阵CCD传感器采集的纱线直径数据,探究在大数据容量下较高精度地实现纱线直径预测具有重要意义。
转载注明来源:https://www.xzbu.com/1/view-15427026.htm