基于多源检测器的交通流数据融合方法研究

发布时间：2018-04-22 19:29

本文选题：数据融合 + 粒子群优化算法　；参考：《北京交通大学》2017年硕士论文

【摘要】：智能交通系统作为解决城市交通拥堵问题的有效途径得到了众多城市的一致认可。诸如车牌识别、RTMS检测器等交通流参数智能感知设备的不断丰富,在很大程度上促进了交通管理的智能化发展。然而,不同设备获取的交通流参数信息在格式、种类等方面存在一定的差异,同时单一检测设备往往存在信息缺失、信息错误等质量问题,这将导致决策失误。因此,如何实现多种设备信息的互补校验,从而提高交通流参数信息的精确度和可靠性尤为重要。本文基于北京市二环快速路交通流实测数据,提出了面向区间实时车速估计的多源异质交通流数据融合方案。首先,基于多源动态交通流实测数据分析了数据的原始属性和交通流时空相关特性;其次,通过对交通领域现有融合方法的优缺点进行对比分析确定了本文的融合方法;再次,针对交通流参数的数据质量问题,依照时间点漂移数据、缺失数据、冗余数据及错误数据的顺序分别进行识别与修正研究,采用动态阈值识别法与交通流机理识别法相结合的方式识别错误数据,根据缺失数据个数的不同选取不同的方法进行修补,同时提出了基于改进埃特金插值的错误数据修正方法,并对修正后的数据进行卡尔曼滤波降噪处理;最后,基于神经网络建立了小波神经网络融合模型与自适应变异粒子群优化小波神经网络融合模型,通过仿真实现并引入MAE、MAPE、RMSE、MSPE、EC与LSE等误差指标对模型进行评价,结果表明本文建立的两种模型的融合精度要明显优于传统的BP神经网络,精度均达到90%以上,且误差离散程度较小,同时对比了样本集处理前后的小波神经网络融合模型的融合效果,验证了样本处理的重要性。
[Abstract]:As an effective way to solve the problem of urban traffic congestion, Intelligent Transportation system (its) has been unanimously recognized by many cities. The abundance of intelligent sensing devices for traffic flow parameters, such as license plate recognition and RTMS detector, promotes the intelligent development of traffic management to a great extent. However, there are some differences in the format and type of traffic flow parameter information obtained by different equipments. At the same time, the single detection equipment often has quality problems such as missing information, information error and so on, which will lead to decision-making errors. Therefore, how to realize the complementary calibration of various equipment information and improve the accuracy and reliability of traffic flow parameter information is particularly important. Based on the measured traffic flow data of Beijing second Ring Expressway, a multi-source heterogeneous traffic flow data fusion scheme for interval real-time speed estimation is proposed in this paper. Firstly, based on the measured data of multi-source dynamic traffic flow, the original attributes of the data and the space-time correlation characteristics of the traffic flow are analyzed. Secondly, the fusion method of this paper is determined by comparing the advantages and disadvantages of the existing fusion methods in the field of transportation. Aiming at the problem of data quality of traffic flow parameters, according to the order of time drift data, missing data, redundant data and error data, the recognition and correction of traffic flow parameters are studied respectively. The method of dynamic threshold recognition and traffic flow mechanism recognition is used to identify the wrong data. According to the number of missing data, different methods are selected to repair the error data. At the same time, an error data correction method based on improved Etkin interpolation is proposed, and the modified data is processed by Kalman filter noise reduction. Finally, Wavelet neural network fusion model and adaptive mutation particle swarm optimization wavelet neural network fusion model are established based on neural network. The results show that the fusion accuracy of the two models is obviously better than that of the traditional BP neural network, and the accuracy is more than 90%, and the degree of error dispersion is small. At the same time, the fusion effect of wavelet neural network fusion model before and after sample set processing is compared, and the importance of sample processing is verified.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U491

【参考文献】