urban road, traffic bottleneck, warning, short-term traffic flow prediction, dimension reduction, headway, joint tree method, reinforcement learning algorithm, signal coordination control
Traffic congestion prevention and control are the long-term problems in urban road traffic management. Traffic bottleneck is often the starting point of traffic congestion. Therefore, it is of great significance to study traffic bottlenecks in urban road network. Traffic bottlenecks warning is mean to identify traffic nodes in urban road network that may develop into traffic bottlenecks before traffic congestion, by analyzing the current traffic condition, determining the development trend, and studying the cause of traffic bottlenecks. Traffic bottlenecks warning provide a basis for decision-making on formulating and adjusting traffic management measures, and achieve the purpose of delaying or even avoiding traffic jams. Intersection is the convergence point of multi-directional traffic flow, resulting in low efficiency of traffic flow in all directions. Signal control is the most effective way to ensure traffic efficiency and reduce traffic conflicts at intersections, and the coverage in urban road network is wide. Therefore, signal intersections will serve as the main target for traffic bottlenecks warning in urban road networks. In order to provide theoretical and technical support for urban road network traffic bottleneck warning, based on the requirement of traffic bottleneck warning research, the paper takes the signalized intersection as the main research object, and studies the traffic bottleneck warning process and key support technologies.
Based on the current research on traffic bottleneck and the demand of traffic bottleneck management, this paper puts forward the traffic bottleneck warning process in urban road network and key technical requirements of traffic bottleneck warning, and analyzes the short-term traffic flow prediction of urban roads, the characteristic parameters of signalized intersections and control methods, regional coordination of signal control and other key technologies to carry out research. The specific research contents are as follows.
(1) Based on the analysis of traffic bottlenecks characteristics and traffic management requirements, this paper puts forward the traffic bottleneck warning process and key technical demand in urban road network.
By analysing traffic bottleneck’s style and basic characteristics, taking urban road traffic bottleneck management as a guideline, the traffic bottleneck warning process is divided into four parts, which are traffic information collection and analysis, potential traffic bottleneck identification, potential traffic bottleneck incentives analysis and traffic bottlenecks warning information release. Among the key technologies that support traffic bottleneck warning, short-term traffic flow prediction technology is an important tool for potential traffic bottlenecks identification, traffic characteristics analysis technology of signalized intersections is the basis for potential traffic bottleneck identification method and signal control efficiency improvement, coordinated signal control is an important means of analyzing and improving potential traffic bottlenecks. Although the above key technologies have achieved some results, they still have much room for improvement. This paper will further study and improve the above technologies.
(2) Based on the spatiotemporal correlation of traffic flow data in urban road network, a dimensionality reduction method of urban road traffic flow temporal-spatial data is proposed to improve the accuracy of the existing short-term traffic flow prediction model.
Existing short-term traffic flow prediction models are mainly applied to the prediction of highway traffic flow. The direct application to urban road is difficult to adapt to the fluctuation of urban road traffic flow, and the prediction accuracy is low. Based on the spatial and temporal correlation of urban road traffic flow, this paper proposes a dimensionality reduction method for urban road traffic flow in order to improve the accuracy of applying existing short-term traffic flow prediction models in urban road. Qualitative analysis is used for primary selection of road sections, the multi-dimensional scale is used to cluster and group primary selected road sections, the correlation analysis is used to decide the final section. The effects of the proposed method on the forecasting results are verified by using BP neural network and multivariate linear regression model respectively. The results show that the proposed dimensionality reduction method can improve the prediction accuracy and speed of urban road short-term traffic flow forecasting model.
(3) Based on the analysis of the sensitivity and calibration accuracy of characteristic parameter of the signalized intersection, the paper proposes the calibration method and the signal control method of the signalized intersection by studying the distribution of the headway of the queued vehicles at the signalized intersection.
The signal loss time, saturation flow rate and headway are the main characteristic parameters of the signalized intersection. Through the analysis of the sensitivity and calibration accuracy of the characteristic parameters, the drawback of existing calibration methods are obtained. The method to calibrate the characteristic parameters through the distribution of the headway is advised. Through the study of the headway distribution, it can see that the headway distance at a fixed position obeys the logarithmic normal distribution and the headway at a continuous queue position obeys a logarithmic distribution, and it is not advisable to calibrate the headway of each queuing position by a fixed quantile. Based on the distribution of headway, the paper discusses the calibration method of signalized intersection characteristic parameters and the method of signal control which does not take saturated flow rate and total loss time as variables, and proposes the optimal green light time control method, the calculation method of dynamic capacity at intersections, and the signal timing algorithm that not relies on traffic capacity and signal loss time.
(4) The regional traffic signal coordination control is carried out by combining the joint tree method and reinforcement learning algorithm. The applicability of the algorithm to the traffic signal control is improved by the basic theory optimization. The requirement of traffic bottleneck warning on signal coordination control is to be satisfied by the application function optimization.
In order to meet the demand of traffic bottleneck warning, this paper proposes a coordinated tree control method and an enhanced learning algorithm for coordinated control of regional traffic signals. However, existing researches are still insufficient in information transmission, parameter settings, action selection strategies, maximum spanning tree analysis, and single intersection operation; it is difficult to meet traffic bottlenecks warning. Based on the existing research, this paper proposes to optimize the regional signal coordination algorithm from the aspects of optimization of basic theory and application of functional optimization. Optimization of basic theory includes optimization of core parameters, optimization of basic rules and optimization of JTA information delivery modes. Optimization of application functions includes algorithm optimization based on the effect of single intersection operation and reward function optimization based on identification of traffic bottlenecks. The test environment is constructed to validate the improved algorithm. The results show that the improved algorithm not only can better meet the basic needs of traffic signal control, but also meet the traffic bottleneck warning.
(5) Based on the actual urban traffic environment, the application of the research results is analyzed.
Traffic simulation environment is constructed based on part of the road network and traffic flow data of Changshu City. Headway distribution and dynamic traffic capacity at signalized intersections is verified by using headway data of various lanes. Taking traffic simulation environment as a platform, the potential traffic bottleneck identification method based on real-time traffic flow status is applied, and different evaluation indexes are used to identify potential traffic bottlenecks. The method of dimensionality reduction of traffic flow spatial data for short-term traffic flow prediction is applied to verify the effect of dimension reduction method on the prediction accuracy and speed of urban short-term traffic flow. Based on the identification of potential traffic bottlenecks, the reason of the traffic bottleneck is analyzed by using the regional signal coordination and reinforcement learning algorithm based on the joint tree method.