基于双层混合集成的自动驾驶汽车故障检测  

Fault detection of autonomous vehicle based on bi-layer hybrid ensemble

作　　者：闵海根 雷小平 李杰 童星 吴霞[1,3] 方煜坤 MIN Haigen;LEI Xiaoping;LI Jie;TONG Xing;WU Xia;FANG Yukun(School of Information and Engineering,Chang'an University,Xi'an 710064,Shaanxi,China;Collaborative Innovation Center for Western China Traffic Safety and Intelligent Cooperative Control,Xi'an 710021,Shaanxi,China;The Joint Laboratory for Internet of Vehicles,Ministry of Education-China Mobile Communications Corporation,Chang'an University,Xi'an 710021,Shaanxi,China;Shandong Hi-Speed Information Group Co.Ltd.,Jinan 250014,Shandong,China)

机构地区：[1]长安大学信息工程学院,陕西西安710064 [2]西部交通安全与智能协同控制省部共建协同创新中心,陕西西安710021 [3]长安大学“车联网”教育部-中国移动联合实验室,陕西西安710021 [4]山东高速信息集团有限公司,山东济南250014

出　　处：《山东大学学报（工学版）》2022年第6期30-40,共11页Journal of Shandong University（Engineering Science）

基　　金：国家重点研发计划项目(2021YFB2501205);国家自然科学基金青年项目(61903046);陕西省自然科学青年基金(2022JQ-663)。

摘　　要：针对单一故障检测算法难以学习到数据样本全部特征的问题,提出基于双层混合集成的无监督自动驾驶汽车故障检测方法。使用非全连接的自动编码器作为基学习器构建第1层同质集成框架——集成自动编码器,分析和选择包含集成自动编码器、一类支持向量机、孤立森林和局部离群因子的基学习器构建第2层异质多模型集成框架,学习自动驾驶汽车正常传感器数据特征;提出基于自动编码器的投票集成方法,实现基学习器特征的降维和编码融合;通过sigmoid函数映射计算故障概率并对数据是否故障进行判断。试验结果表明,提出的双层混合集成故障检测方法性能优于基学习器算法,F_(1)指标提高了9%~40%,G指标提高了2%~28%,该故障检测方法可有效实现自动驾驶汽车故障检测。Aiming at the problem that it was difficult for a single fault detection algorithm to learn all the features of the data samples,an unsupervised autonomous vehicle fault detection method based on bi-layer hybrid ensemble was proposed.The non-fully connected autoencoder was used as the base learner to build the first layer homogeneous ensemble framework—ensemble autoencoder.In order to further improve the fault detection ability,the base learner including ensemble autoencoder,one-class support vector machine,isolation forest and local outlier factor were analyzed and selected to build the second layer heterogeneous multi model ensemble framework and learn the normal sensor data characteristics of autonomous vehicle.A voting ensemble method based on autoencoder was proposed to realize the dimensionality reduction and coding fusion of base learner features.The fault probability was calculated by sigmoid function,and whether the data was at fault was judged by fault probability.The experimental results showed that the performance of the proposed bi-layer hybrid ensemble fault detection method was better than the base learning algorithm,which was improved by 9%-40% in F_(1) index and 2%-28% in G index.The fault detection method based on autoencoder voting could effectively realize the fault detection of autonomous vehicles.

关 键 词：自动驾驶汽车 故障检测 集成学习 自动编码器 无监督学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]