摘要
随着传感器和监测技术的快速发展,大量工程装备运行数据被记录和保存下来,为提升装备设计、分析和运行水平提供了数据基础。这些数据内部存在的异常值会对数据建模与分析产生严重干扰。为此,文中提出一种基于模糊聚类与多项式回归的异常值识别与鲁棒建模方法,通过多项式回归来刻画数据属性之间的关联关系,采用模糊聚类计算数据的隶属度信息判断数据是否为异常值,并基于非异常值数据建立预测模型,最终实现数据的鲁棒建模。数值案例、隧道掘进机和联合收获机监测数据试验结果表明,提出方法能够准确识别出数据中的异常值,并显著提升数据建模精度。
With the rapid development of sensors and monitoring technologies,a massive amount of engineering equipment data has been recorded,providing a data basis for improving equipment design,analysis,and operation levels.However,these data often contain outliers,which can degrade the accuracy of data modeling.To this end,this paper proposes an outlier identification and robust modeling method based on fuzzy clustering and polynomial regression.By employing polynomial regression to capture the correlations among data attributes and utilizing fuzzy clustering to calculate the degree of membership of data points,the proposed method effectively identifies outliers.Furthermore,a prediction model is built based on the non-outlier data to improve its robustness.Experimental results on numerical cases,tunnel boring machine data,and combine harvester monitoring data demonstrate that the proposed method achieves accurate identification of outliers and significantly enhances the accuracy of the prediction model.
作者
石茂林
张增磊
张卫东
谈莉斌
钟良意
赵梦晨
SHI Maolin;ZHANG Zenglei;ZHANG Weidong;TAN Libin;ZHONG Liangyi;ZHAO Mengchen(Key Laboratory for Theory and Technology of Intelligent Agricultural Mechinery and Equipment,Jiangsu University,Zhenjiang 212013;International Science and Technology Cooperation Base for Intelligent Equipment Manufacturing in Special Service Environments,Anhui University of Technology,Ma’anshan 243032;Wuhan Second Ship Design and Research Institute,Wuhan 430064)
出处
《机械设计》
CSCD
北大核心
2023年第S02期45-50,共6页
Journal of Machine Design
基金
江苏大学人才引进启动资金(20JDG068)
江苏省自然科学基金资助项目(BK20210777)
中国博士后科学基金面上项目(2022M711388)
省部共建现代农业装备与技术协同创新中心资助(XTCX2014)
关键词
异常值识别
鲁棒建模
模糊聚类
多项式回归
outlier detection
robust modeling
fuzzy clustering
polynomial regression
