Dynamic Analysis And Applications Of Traffic States In Urban Road Network Driven By Electronic Maps

Because of the relatively complete social services and huge employment demand,people quickly gather in cities.The developing speed mismatch between residents' travel demand and the urban transportation supply capacity makes traffic congestion a common problem in cities.In order to improve the systematic and scientific nature of congestion management,it is of great significance to study the dynamic characteristics and application of urban traffic conditions.The paper takes electronic real-time traffic condition maps as the research data.The data integrates information of the original traffic flow and geographic information,in which pixel colors record the real-time traffic conditions of the road sections,which better retains the overall and continuous temporal and spatial characteristics of the city's traffic state.By mining this map series data,the paper attempts to solve the following three problems.First,it quantifies the influence of each road segment on the congestion degree of the urban road network and studies the way to optimize the management of the road sections with great influence.Then,it quantitatively evaluates the evolution of urban traffic status covering the road network in peak hours to evaluate the implementation effect of traffic restriction policies.Finally,it allocates vehicles from origin to destination dynamically to avoid congested road sections,considering the road states in the real-time traffic condition map and the queuing in related service stations.The online navigation system does the guidance function.Finally,the main research contents of this thesis are the three aspects listed following.(1)Identify the key congestion road based on the bijective decision theory.We grid the electronic real-time traffic condition map,make the pixels as dimensions and its color as the feature values of the dimensions,convert it into high-dimensional massive sequence data.Based on analyzing the number distribution of the three colors' pixels,it creates one three-segment function to category the urban traffic states.Then,it constructs one expert system based on the bijective theory.We improve the feature reduction algorithm BSSReduce to reduce the feature dimensions of high-dimensional vectors and lock the key congested road sections of urban traffic.This method can find 70 points from10,000 pixels and accurately locate the 60*60 square meters' road network grid.Compared with the traditional method,the cost is lower,the coverage is higher,and the scope is more accurate.The research can help urban traffic operators and managers in the complex urban road network to quickly locate the road sections that have a great influence on the road network traffic state,and better carry out road network planning and road section traffic management.(2)Urban traffic states' evolution analysis based on convolutional autoencoder.With the sequence of electronic real-time traffic condition map,based on the autoencoder model,we do representation learning and get the feature vectors in each map's latent space.We propose a low-cost,large-coverage,and quantitatively accurate method to mine the potential spatial characteristics of the road-network traffic state sequence.This method can capture the overall urban traffic during peak hours.Compared with the traditional histogram,the state change process captures richer characteristics of the temporal and spatial changes of the traffic state.The research can help urban traffic management departments to better grasp the changes in road network status and verify the effects of related traffic management measures.(3)Dynamic vehicle allocation strategy considering the network traffic states.Combining the known road condition data,we try to control the vehicle behaviors.With the help of an online navigation system,our method can guide vehicles to avoid congested areas actively.Taking the charging behavior of electric vehicles as a case scenario,we construct a congestion game model to describe the interaction process between vehicles and easily congested resources,and propose an adaptive learning method to match the electric vehicles and charging stations to avoid congestion.Comparing with the traditional genetic algorithm,the performance of this method has no relationship with the initial value,and the convergence is more rapid.The research can further optimize the drainage function of the online navigation system.