面向CIM和动态交通分析的多源异构数据融合技术研究  

Research on multi-source heterogeneous data fusion technology for CIM and dynamic traffic analysis

作　　者：王志成 张玉一 巴天星 常燕燕 WANG Zhicheng;ZHANG Yuyi;BA Tianxing;CHANG Yanyan(Shenyang Planning&Design Institute,Shenyang 110004,China)

机构地区：[1]沈阳市规划设计研究院有限公司,辽宁沈阳110004

出　　处：《电子设计工程》2024年第8期68-72,共5页Electronic Design Engineering

摘　　要：针对当前智慧交通系统仅能对少量指标进行分析的问题,文中基于CIM三维实时交通数据以及多源数据系统提出了一种动态城市交通分析算法。该算法由多尺度DCNN和Bi-LSTM模型组成,其中多尺度DCNN模型可以对CIM三维交通数据进行训练,从而获得实时的交通特征信息。同时由于加入了残差网络,使得模型具有更好的全局视野特征。通过Bi-LSTM对与交通相关的多源数据进行训练,以得到数据的时序特征,由Softmax网络对多源数据特征实现融合并获得最终结果。实验测试结果表明,所提算法的图像识别性能与多源数据分类性能在对比算法中均为最优,且识别准确率可达87%,证明其具备良好的实时交通状态识别能力。Aiming at the defect that the current intelligent transportation system can only analyze a few indicators,this paper proposes a dynamic urban traffic analysis algorithm based on CIM 3D real-time traffic data and multi-source data system.The algorithm is composed of DCNN and Bi LSTM models.DCNN model can train CIM three-dimensional traffic data to obtain real-time traffic feature information.At the same time,because of the residual network,the model has better global view characteristics.BiLSTM is used to train multi-source data related to traffic to obtain the time series characteristics of data,and then Softmax network is used to fuse multi-source data characteristics to obtain the final result.The experimental results show that the image recognition performance and multi-source data classification performance of the proposed algorithm are both the best in the comparison algorithm,and the accuracy rate can reach 87%,indicating that it has good real-time traffic state recognition capability.

关 键 词：多源数据 深度卷积神经网络 长短时记忆神经网络 残差网络 动态交通分析 CIM系统 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN929.5[自动化与计算机技术—计算机科学与技术]