Research On Traffic Speed Prediction And Traffic Guidance In Urban Road Network

With the rapid development of the economy and improvement of people’s living standards,the number of vehicles on urban roads continues to increase,bringing convenience to travel and serious traffic congestion at the same time.Intelligent Transportation System(ITS)is a kind of efficient transportation management system that can effectively reduce traffic congestion,facilitate in advance,and utilize existing transportation infrastructure.Traffic flow prediction method and dynamic traffic guidance system are the core of ITS.Currently,traditional machine learning methods and deep learning methods are widely used for traffic flow prediction,and the accuracy of their prediction results is affected not only by the prediction model but also by the quality of traffic data.However,in the open environment,the traffic flow data collected by ITS generally have quality problems such as incompleteness,abnormality,and noise.To effectively deal with lowquality traffic flow data and improve the accuracy of traffic flow prediction,it is necessary to carry out research on traffic flow prediction methods for noisy data.Meanwhile,the traffic guidance technology using road traffic flow have been relatively mature,but the existing results rarely consider the differences in traffic flow between different lanes of the same road,causing potential deviations in traffic guidance and frustrating navigation user experience.Therefore,considering the current mature vehicle navigation technology,it is necessary to carry out the key problem of dynamic traffic guidance time information based on the heterogeneity of multi-lane traffic flow.To solve the key problems such as accurate prediction of traffic speed with different characteristics of noisy data and differential utilization of dynamic traffic guidance information,the specific research contents are as follows.(1)Traffic speed prediction method considering the overall characteristics of data noise.From the overall characteristics of data noise,a robust prediction method is proposed to avoid or reduce the impact of noisy traffic speed data.Based on the robust model of traditional extreme learning machine,a multi-extreme learning machine framework is proposed to improve the ability to fit complex changes in traffic data.Further,assuming that the overall characteristics of the complex noise in low-quality traffic data obey a mixture distribution,this dissertation proposes the robust multi-extreme learning machine prediction model based on infinite mixture of Gaussians(MoG)and infinite mixture of Student’s-t distributions.Finally,under the sparse Bayesian framework,the model can adaptively learn the model weights,output weights,parameters,as well as the number of distributions in the mixture distribution.The effectiveness of the model is verified on artificial datasets,benchmark datasets and real traffic speed datasets.It is demonstrated that the proposed model outperforms traditional comparative models in low-quality data environments.(2)Traffic speed prediction method considering individual characteristics of data noise.Based on the individual characteristics of data noise,the dissertation proposes a traffic speed prediction method based on multi-scale feature information and complex individual characteristics of data noise.A combination model(CBiLSTM)is established by applying various deep learning methods for traffic speed prediction,and a multi-scale feature extraction module(MCBiLSTM)based on CBiLSTM is proposed.MoG with different parameters is used to portray the variability of noise individuals in traffic flow speed data,and a heterogeneous MoG-based loss function is derived to train the parameters of MCBiLSTM.On this basis,a traffic speed prediction model(MCBiLSTM-MoG)based on multi-scale information and heterogeneous MoG-based loss function is constructed.The model can directly output not only the traffic speed point prediction,but also the traffic speed interval prediction.On real traffic speed data,it is verified that the proposed model has better performance in both point and interval prediction performance of traffic speed on different roads.(3)Dynamic traffic guidance considering the heterogeneity of traffic speed of multi-lane.From the heterogeneity of multi-lane traffic flow speed,the guidance time information is updated to solve dynamic traffic guidance problem based on heterogeneous multi-commodity dynamic network flow.The Quickest Heterogeneous Multi-commodity Dynamic Network Flow(HeteroMDF)problem based on heterogeneous traffic speed is proposed.The dissertation defines the heterogeneous time expanded network and complete the proof of the pseudo-polynomial time algorithm for HeteroMDF problem in the 2-lane case.On this basis,the dissertation introduces a new capacity constraint to ensure that feasible static flows in the heterogeneous time-expanded network can be transformed into feasible dynamic flows,and complete the proof and computation of the fully polynomial time approximation scheme for HeteroMDF problem in the multi-lane case.Then the traffic guidance based on heterogeneous multicommodity dynamic network flows is proposed.The experiments verify the effects of road and lane traffic speed on the computed traffic impedance and discover the differences of dynamic traffic guidance using road traffic speed and lane traffic speed.