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

Accurate and timely detection of highway traffic congestion is the premise for relevant departments to take management measures to prevent secondary accidents and reduce economic losses.The traditional traffic congestion detection research is mostly based on a single data source,due to the limitation of working principles and methods,the detection based on a single data source is difficult to fully reflect the traffic state of the highway.With the abundance of highway data collection methods,some highways now have a variety of traffic data sources.By comprehensive utilization of the complementarity and redundancy among multiple data sources,the accuracy of traffic congestion detection can be effectively improved through multi-source data fusion.However,the existing data fusion method is not fully applicable to highway traffic congestion detection,and its detection rate needs to be improved.Based on the actual environment and existing conditions of highways,the research on the highway segment traffic congestion detection method based on multi-source data fusion has important theoretical and practical significance for traffic managers to implement traffic control and traffic travelers to plan travel plans.In this dissertation,by analyzing the correlation between the three parameters of the vehicle detector data,a three-dimensional Mc Master algorithm for detecting traffic congestion in cross-section is proposed.In order to make full use of toll data,the SND-GMM algorithm is proposed to detect traffic congestion on road segments.According to the characteristics of the detection form and range of multi-source traffic data,the decision-level fusion of the detection results of single data sources is implemented to realize the detection of traffic congestion on highway segments.The main content of the dissertation includes:(1)A three-dimensional Mc Master algorithm based on vehicle detector data is proposed.To solve the problem that the density of vehicle detector is not enough to apply double section detection algorithm and the detection rate of single section detection algorithm is low,in this dissertation,the traditional Mc Master algorithm is extended to the three-dimensional parameter space by analyzing the correlation between the three parameters of speed,flow,and occupancy in the data of the vehicle detector.The actual data on Chongqing highway show that the 3D Mc Master algorithm can significantly improve the detection rate of traffic congestion in section.(2)The SND-GMM algorithm based on toll data and checkpoint data is proposed.In view of there is less toll data between adjacent toll stations,the checkpoint data is used to supplement toll data.The travel speed of the road segment is used to characterize the traffic state of the road segment.Aiming at the problem that the traditional SND algorithm has too many false and missed detections and cannot continuously detect traffic congestion,this dissertation reselects the discriminant feature,use the gaussian mixture model for clustering analysis,and then the clustering results are used to detect road congestion.Verification through actual data shows that the SND-GMM algorithm can effectively detect traffic congestion on road segment.(3)The multi-source data decision fusion method based on FFM model is proposed.The similarity between traffic congestion detection based on multi-source data decision fusion and CTR prediction is analyzed,and then the FFM model with outstanding performance on the CTR prediction is introduced into the research of traffic congestion detection to achieve decision-level fusion of multi-source traffic data.The experimental results show that the multi-source data decision fusion method based on the FFM model can be applied to traffic congestion detection,which effectively improves the detection rate of road traffic congestion.In summary,the traffic congestion detection method for highway segements based on multi-source data decision fusion proposed in this dissertation can improve the detection rate of road congestion on the basis of not increasing existing costs,has high practicability and feasibility,which can provide timely,reliable and predictable decision support for managers and travelers.