| 引用本文: | 赵慧英,钱大琳,张博,范爱华.危险货物道路运输车辆出行链活动类型识别[J].哈尔滨工业大学学报,2019,51(9):193.DOI:10.11918/j.issn.0367-6234.201811079 |
| ZHAO Huiying,QIAN Dalin,ZHANG Bo,FAN Aihua.Activity type recognition of trip chain for hazmat road transportation vehicle[J].Journal of Harbin Institute of Technology,2019,51(9):193.DOI:10.11918/j.issn.0367-6234.201811079 |
|
| 摘要: |
| 针对危险货物运输(危货)车辆出行链上停留节点活动类型的识别问题,提出基于高斯混合模型-隐马尔科夫模型(GMM-HMM)的活动类型识别方法. 对车辆GPS数据构建基于决策树的车辆起停检测模型,提取出行链活动节点,并通过D-OPTICS算法对活动节点聚类得到活动热区；根据活动节点的个体特征、所在出行链的亲属特征和所处热区的群体特征进行多尺度特征体系构建,通过因子分析进行降维处理；利用GMM-HMM构建危货车辆活动类型识别模型,通过Baum-Welch算法进行参数估计,并使用Viterbi算法解码隐藏状态得到出行链各活动节点的类型识别结果. 在小规模实际活动数据集上直接验证所提方法的正确性,还结合活动节点的POI类别,间接评估所提方法对大规模GPS数据的车辆活动类型识别效果. 实验结果表明:在9种活动类型识别任务中,基于GMM-HMM的出行链活动类型识别方法的活动识别率超过80%. 识别结果可用于分析活动行为模式,及时发现异常活动,为危险品运输监管提供有效的决策支持. |
| 关键词: 危险货物运输 GPS数据 出行链活动类型 多尺度特征体系 隐马尔科夫模型 高斯混合模型 |
| DOI：10.11918/j.issn.0367-6234.201811079 |
| 分类号:U492.3+36.3 |
| 文献标识码:A |
| 基金项目:北京市交通委科研项目(T18100100); 河南省交通运输厅科技项目(2019G-2-8) |
|
| Activity type recognition of trip chain for hazmat road transportation vehicle |
|
ZHAO Huiying1,QIAN Dalin1,2,ZHANG Bo1,FAN Aihua1
|
|
(1.School of Traffic and Transportation, Beijing Jiaotong University, Beijing 100044, China; 2.Key Laboratory of Transport Industry of Big Data Application Technologies for Comprehensive Transport(Beijing Jiaotong University), Beijing 100044, China)
|
| Abstract: |
| This paper proposes a method based on Gaussian-Mixture-Model Hidden Markov Model (GMM-HMM) to recognize the activity-node type of trip chain for hazardous-materials (hazmat) transportation vehicle. The GPS data of vehicle were pre-processed to identify the activity nodes of the trip chains using a Decision-Tree based move-stop detection algorithm. Then the activity nodes were grouped into the activity hotspots by a dropout-based OPTICS (D-OPTICS) algorithm. A multi-scale feature system was constructed according to the individual features of the activity nodes, the relative features of the corresponding trip chains, and the group features of the related hotspots. These feature vectors were further transformed into low-dimensional vectors using Factor Analysis method. Finally, a GMM-HMM based activity type recognition model for hazmat transportation vehicles was built where Baum-Welch algorithm was used for parameter estimation, and Viterbi algorithm for decoding the hidden state to obtain the recognition results of the activity-node type of trip chains. Not only the accuracy of the proposed method was directly verified based on the small-scale real-activity dataset, but also the effectiveness of the proposed method on the activity type identification of large-scale GPS data was evaluated indirectly using the Point-of-Interest (POI) information. The results demonstrate that the identification rate of the GMM-HMM based activity type recognition method was more than 80% in the task of nine-type activity recognition. The recognition results can help analyzing the activity behavioral patterns, discovering the abnormal activities, and providing effective decision-making support for hazmat transportation supervision. |
| Key words: hazardous-materials transportation GPS data activity type of trip chain multi-scale feature system Hidden Markov Model Gaussian Mixture Model |
