Research On Traffic Flow Forecasting Of Correlated Roads Based On Road Network Topology

Intelligent transportation system(ITS)is considered to be an important means to solve the existing traffic,environmental and social problems.Efficient,accurate and comprehensive traffic data collection and processing ability is the premise of ITS,which directly affects the accuracy of analysis and decision-making.However,at present,most of the traffic devices are distributed on key roads such as urban expressway,primary or secondary trunk roads,which cannot cover the whole road network.As a result,the traffic data of some roads in the road network cannot be detected so that become the dead zone of traffic information collection system.Therefore,it is of great significance to obtain real-time traffic flow data of the whole road network through a limited number of traffic devices.The existing traffic flow forecast model of correlated road needs detailed OD data and other prior information,and establish the mathematical relationship between the known traffic road and other unknown correlated roads.The modeling process is too complex to guide the layout optimization of traffic devices.Therefore,the traditional multi-section real-time traffic forecast method needs to be improved.In order to solve the above problems,this paper considers the influence of road network topology on the flow correlation,and divides the road sections according to the flow correlation.The real-time prediction model driven by traffic flow data of each correlated road sections is established,which avoids the problem of complex prior data,and realizes the integrated modeling of the traffic flow from the known section to all other unknown correlated roads in the same section.Therefore,the modeling process is more concise.The main research contents are as follows:(1)Based on the summary of Graph Theory and Complex Network Theory,the dualmethod method is used to construct the road network topology structure.Five road network topological structure characteristic indexes affecting the flow correlation are extracted,including the Number of Lane reflecting the physical characteristics of the road section itself,the Shortest Path Length reflecting the relationship between the road sections,and Degree,Closeness and Betweeness reflecting the importance of the road section in the road network.(2)Qualitatively analyze the spatial correlation of different roads of traffic in the urban road network,and clearly define some terms such as " correlated roads" and "key road".The correlation strength of multi-section traffic data is quantitatively analyzed by MDS,then these roads are divided into different sections according to the correlation strength,which provides a basis for multi-section traffic forecast research.(3)For each section of correlated roads,build a MLP neural network traffic prediction model with structure(6,p,q,1).By analyzing the influence factors of multisection traffic correlation,the model input layer variables are determined to be Number of Lane,Degree,the Shortest Path Length,Closeness and Betweeness of the key road;the output layer variables are the traffic of those roads which need to be forecasted;the training is performed using the BP algorithm,by the way,determine the optimal number of hidden neurons(p)for each model during training.The traffic forecast model is applied to a regional road network in Chongqing City,which proves that the model has high accuracy and feasibility,so that can be used to guide the layout of traffic devices in urban road network.