Big Data-driven Urban Road Network Traffic Status Prediction And Pattern Analysis

With the increase in people's travel needs,urban traffic congestion is getting worse.How to effectively alleviate urban traffic congestion and improve the efficiency of urban road traffic has been a research hotspot in academia.Real-time and accurate traffic flow prediction is an important prerequisite for urban traffic control.Decomposing regional coordinated control into coordinated control of several key traffic routes,and performing traffic state prediction and pattern analysis on key routes in the urban road network can effectively simplify traffic control algorithms and help urban traffic control personnel to propose suitable solutions to alleviate urban traffic congestion,thereby improving the efficiency of urban road network traffic operations.This thesis focuses on traffic state prediction and pattern analysis of urban road networks driven by big data.The main research contents and contributions are organized as follows:1)To solve the problem that a single forecasting model cannot take into account both the generalization ability and the processing of dynamic real-time data,a short-term traffic flow combined forecasting model is proposed.First,we perform traffic flow aggregation on the sydney coordinated adaptive traffic system data.Then,through multidimensional scaling and the proposed spatio-temporal feature selection(STFS)algorithm,feature selection is performed on the forecasting model to determine the best input features of the kalman filter(KF)model and support vector regression(SVR)model.Finally,a combined forecasting model is proposed,which linearly weights the predictive results of the KF model and the SVR model after feature selection.Simulation results show that the STFS algorithm proposed in this thesis can improve the performance of single forecasting model,and the predictive accuracy of combined forecasting model is the highest.2)Aiming at the traffic flow deduction of urban road network without real-time data of traffic flow,a traffic flow simulation deduction model based on multi-model combination is proposed.First,we calculate the historical average flow of each road section in the road network,and save the calculation result to the database.Then,long short term memory model is used to predict the traffic flow of each road section in the road network,and state transition probability of each road section is estimated in the road network through the markov chain,and the results of the predictive traffic flow and state transition probability results are saved to the database.Finally,an improved KF model is proposed to deduce the traffic flow of the road network.The simulation results show that the simulation deduction algorithm proposed in this thesis can be used for the deduction of urban traffic flow.3)In order to improve the efficiency of area control,a method of traffic pattern analysis and traffic state prediction for key routes in the road network is proposed.First,the concepts of correlated routes,correlated route chains,and correlated route set are defined,and a route correlation degree calculation model that takes into account the non-uniformity of route flow is proposed and judgement conditions are given to determine the correlated route set in the region.Next,we use the self organizing map algorithm to classify traffic states on correlated route chains and determine the optimal number of traffic states by the Dunn index.Then,traffic patterns on correlated route chains are analyzed.Finally,the SVR model after feature selection is used to predict traffic states on correlated route chains.The simulation results show that among the number of attempted traffic state clusters,the optimal traffic state number of correlated routes is 5.Furthermore,the predictive mean absolute percentage error of traffic states for lanes on the correlated route chain is less than 18%,and the performance of traffic state prediction is good.