Adaptive Ramp Metering Control For Urban Freeway Using Large-Scale Data

In recent years,with the rapid development of China's economy and the increasing speed of urbanization process,the number of motor vehicles has exploded,causing frequent urban traffic congestion.Each city actively promotes the construction of expressway systems to alleviate the huge urban traffic load.The urban freeway is the traffic system in which people and vehicles are completely separated and freeway is the viaduct structure.An efficient freeway system can greatly reduce urban traffic congestion.However,the current urban freeway system,as the convergence section of the huge traffic flow,has exceeded the load.Therefore,combining advanced information technology and exploring the more efficient freeway management and control method has attracted the attention of many scholars.Most of research and engineering experience has pointed out that entrance ramp control is the most effective and direct control method for solving freeway traffic congestion[1]However,the existing ramp controller is based on the real-time traffic information of the freeway.The key parameters in traffic management are often set based on experience by the individual or the management department,lacking enough research for the state discrimination,critical traffic congestion threshold,congestion mode,etc.The following problem will happen in the freeway controller:fixed model structure,imprecise ideal threshold,fixed controller parameters and so on.Therefore,based on the large-scale Hangzhou freeway data,after data cleaning and pre-processing,the three traffic factors-speed,flow and density-of each road segment are obtained.Combined with external data such as weather and holidays data,based on the Greedshields model we obtain macroscopic fundamental diagram(MFD)for each road segment.Based on the MFD model,the dynamic congestion threshold of the road segment can be calibrated.By the established congestion threshold calibration model,the historical data is classified into the congestion speed sequences and non-congestion speed sequences.Through the congestion speed sequence and k-means algorithm,thed model clusters congestion speed sequences into different congestion modes and the ramp metering controller parameters are trained for different congestion modes.The specific contributions of this article are as follows:Firstly,this thesis preprocesses the original dataset and obtains three elements-speed,flow,density-for each freeway road segment.Combining the external data such as weather and holidays,this thesis analyzes the internal steady state law and external random factors of traffic speed.On account of the analysis,we establishes a short-time freeway speed prediction model based on the system identification algorithm.It provide the future information for traffic congestion mode prediction and the ramp metering controller can adjust the traffic congestion in advance.Then,based on the historical data of road segment including the flow,density and speed dataset,combined with the weather data,the Greedshields model is used to fit the precise macroscopic fundamental diagram(MFD)for each road segment,and then the critical value of traffic congestion is calibrated.The dynamic ideal density threshold will be set as goal of the ramp controller,so that the ramp controller can follow the change of the traffic state in real time.Finally,in order to further improve the performance of the controller,we clustered a large number of congestion sequences based on the k-means algorithm to obtain a classifier for the congestion mode,so that the adjustment parameters of the controller can be switched with the degree of congestion.When the adjustment parameters are trained for each congestion mode dataset,the system uses the iterative feedback tuning algorithm to automatically iteratively learn and obtain the optimal controller adjustment parameters of each road segment.