Research On Freeway Traffic Congestion Detection Method Based On Multi-source Data Fusion

Timely and accurately detecting the state of the highway traffic congestion is helpful for the traffic control department to detect the traffic accidents in time and take measures to avoid secondary accidents.The actual environment of expressways is complex and changeable.The performance of the traffic congestion detection based on vehicle detectors is limited because of the influence of installation location,working principle,and the uncertainty of a single data source.With the popularization of ETC portal system,a new data source is provided for the characterization of traffic status.Based on the data of vehicle detectors and ETC masts,this thesis studies the highway section traffic congestion detection method based on multi-source data decision-level fusion,which has important research significance and practical value for traffic control and travel planning.In order to make full use of the ETC toll information,the paper proposes a GMM algorithm based on the combination of the characteristics of the average travel speed,flow and density to identify road section traffic congestion;then a SMOTE-ENN-KELM model is proposed combing with the data of vehicle detectors to identify traffic congestion.Finally,combining the characteristics of ETC data,vehicle inspector data and GPS floating car data,a decision-level traffic congestion detection algorithm based on fuzzy comprehensive evaluation of multi-source data is proposed to improve the performance of highway traffic congestion detection.The main contents of the paper include:(1)A GMM expressway congestion discrimination algorithm based on ETC toll data is proposed.As the traditional ramp toll data is only based on the single information of the average travel speed which cannot fully characterize the traffic status of the road section.To solve the problem,the thesis combines the data collected by the existing ETC portal frame,and extract the average travel speed,flow rate and density of the road section to fully characterize the state of road section.In addition,the GMM model is built to realize the detection of traffic congestion on the road section.(2)The SMOTE-ENN-KELM cross-section traffic congestion detection model based on vehicle detector data is proposed.Because of the large distance between highway vehicle detectors,the traditional single-section congestion detection algorithms cannot fully integrate traffic flow data.Thus,the thesis proposes a KELM congestion detection model that combines factors such as speed,flow and occupancy.As for the problem of imbalance between congested and non-congested data in actual situations,the SMOTE oversampling technology is introduced to achieve the balance of the two data.At the same time,the ENN data cleaning technology is introduced to reduce the redundant sample problem caused by oversampling.Therefore,the SMOTE-ENN-KELM congestion detection model is comprehensively constructed and the detection of traffic congestion of vehicle detector road section is realized as well.(3)A fuzzy comprehensive evaluation decision-level fusion algorithm that integrates multi-source data for judging road traffic congestion is proposed.In view of the limited detection performance of each detector,a fuzzy comprehensive judgment decision-level congestion detection algorithm is proposed that combines with the respective characteristics of ETC toll data,vehicle inspection data and GPS floating car data.The mutual complementation and fault tolerance of each algorithm is achieved.In summary,the thesis constructs a set of traffic congestion detection methods based on decision-level fusion of multi-source data.The data of Chongqing expressway section is used for verification.The verification results show that the traffic congestion detection method proposed in the thesis can reduce the congestion false alarm rate and increase the detection rate under the existing equipment deployment compared with the congestion detection based on a single data source.The research results suggest that the traffic congestion detection method based on multi-source data decision-level fusion proposed in the paper is of high engineering practicability and great significance.