摘要
针对目前路域环境感知系统易受路面结构和气候等众多因素影响从而造成感知数据出现异常的问题,对路域环境感知数据异常智能检测与修复问题展开研究,提出一种基于DS-LOF(Difference&Summation-Local Outlier Factor)与GA-XGBoost(Genetic Algorithm-eXtreme Gradient Boosting)的路域环境异常感知数据智能检测与修复方法。以沥青路面温湿度感知数据为实例,首先通过对感知数据进行一阶差分与线性求和计算,构建原始感知数据DS(Difference&Summation)特征向量;然后,基于DS-LOF算法对感知数据进行异常值检测,并与K-means聚类和单类支持向量机算法进行对比分析;其次,以原始感知数据集为基础,并结合异常检测结果,构建路域环境感知数据异常修复数据集;最后基于遗传算法优化XGBoost模型进行数据修复。试验结果表明:GA-XGBoost模型相比于XGBoost模型以及其他机器学习修复模型,其路域环境感知数据修复平均误差最低(M_(AE)=1.2537,R_(MSE)=1.8967),且修复精度最高(R^(2)=0.9448)。对修复前后数据进行稳定性评价,结果表明修复后数据的稳定性评价指标更优,说明修复后数据异常值更少,分布更加稳定。同时设置不同异常占比数据集并对其进行稳定性评价,发现数据集的异常占比越高,数据修复的效果也越明显。提出的路域环境感知数据智能检测与修复模型能够实现对异常数据的智能检测与修复,且能够提升路域环境感知数据质量和稳定性,可为路面性能影响因素分析与衰变规律研究提供可靠的数据支持。
To address the problem of pavement sensing data anomalies caused by many factors,such as pavement structure and climate,research on the intelligent detection and repair of pavement sensing anomalies was conducted.An anomaly detection and repair method for pavement environment sensing data based on the difference and summation local outlier factor(DS-LOF)and genetic algorithm extreme gradient boosting(GA-XGBoost)approaches was developed.Taking the temperature and humidity sensing data of asphalt pavement as an example,first,the difference and summation feature vector of the original sensing data was constructed using first-order difference and linear summation calculation of the temperature and humidity sensing data.Second,the DS-LOF algorithm was used to detect the anomalies of the environment sensing data and compared with the K-means and single-class support vector machine algorithms.Then,an anomaly repair dataset of pavement environment sensing data was constructed based on the original sensing dataset and the anomaly detection results.Finally,the XGBoost model was optimized based on the genetic algorithm for data repair.The experimental results show that the GA-XGBoost model has the lowest average error(mean absolute error of 1.2537,mean squared error of 1.8967)and the highest repair accuracy(R^(2)=0.9448)compared with the XGBoost model and other machine-learning repair models for pavement temperature and humidity sensing data.A stability evaluation of the data before and after the repair was performed,and it was found that the stability evaluation index of the data after the repair was better,indicating that the repair data have fewer outliers and a more stable distribution.Simultaneously,datasets with different anomaly proportions were established,and their stabilities were evaluated.It was found that,the higher the proportion of anomalies in the dataset,the more obvious the important role of data repair.The proposed intelligent detection and repair model of pavement environment sensing data can realize intelligent detection and repair of anomalous data.It can improve the quality and stability of pavement environment sensing data and provide reliable data support for the analysis of the factors influencing pavement performance and the study of the decay law.
作者
孙朝云
裴莉莉
徐磊
李伟
杜耀辉
SUN Zhao-yun;PEI Li-li;XU Lei;LI Wei;DU Yao-hui(School of Information Engineering,Chang'an University,Xi'an 710064,Shaanxi,China;Baidu Online Network Technology(Beijing)Co.Ltd.,Beijing 100193,China)
出处
《中国公路学报》
EI
CAS
CSCD
北大核心
2023年第4期15-26,共12页
China Journal of Highway and Transport
基金
国家重点研发计划项目(2018YFB1600202)
长安大学博士研究生创新能力培养项目(300203211241)
国家自然科学基金项目(52178407)
作者简介
孙朝云(1962-),女,安徽太和人,教授,博士研究生导师,工学博士,E-mail:zhaoyunsun@126.com;通讯作者:裴莉莉(1995-),女,河北邯郸人,工学博士,E-mail:peilili@chd.edu.cn。
