Research On Correlation Of Urban Road Congestion Based On Multi-source Data

In the past 10 years or so,the traffic congestion problem in major cities in China has become more and more serious,and it has become the focus of attention of the people and government departments.The application of Intelligent Transportation System(ITS)helps to alleviate road congestion.The current research work mainly focuses on real-time estimation and prediction of changes in traffic flow based on the data collected by traffic sensors,so as to help the traffic department to make decisions about traffic jams.Correlation analysis of road congestion is one of the important applications.It helps traffic management departments to rationally allocate police force and improve the ability of road resource allocation to effectively alleviate traffic congestion by estimating the degree of time and space correlation of road congestion in the urban road network.In order to improve the accuracy of road congestion correlation estimation in large-scale urban road networks,this paper first proposes a congestion correlation estimation framework based on multi-source traffic data analysis,which is based on road network,vehicles,and geographic points of interest(POI).More than 20 features of traffic information were extracted to calculate the correlation between different road sections and congestion occurrences at different times.Four classifiers including random forest,SVM,decision tree and logistic regression were tried;in order to eliminate the influence of outliers on the estimation accuracy,Proposed a multi-statistical outlier elimination method(MIZ),which combines the median absolute deviation(MAD),interquartile range(IQR)and standard score(Z-score)3statistics,which improves robustness of outlier recognition.In order to solve the problem of imbalance of training data,this paper adopts the idea of model integration,which combines classification models such as gradient boosting decision tree(GBDT),Extreme Gradient Boosting(XGBoost)and adaptive boosting(Ada Boost)to alleviate The impact of data imbalance improves the model recall rate.The data used in the experiment comes from taxi GPS data from August to September in a district in Taizhou City,Zhejiang Province,to estimate the road traffic speed and congestion level every 5 minutes.The GPS data sampling frequency is less than 2 minutes,with a total of 1.7 million data,covering 375 major urban road sections.The POI data is crawled from the traffic map website using crawler technology,a total of 12978 items.The frame accuracy rate based on multi-source traffic data is 84.9%,which basically meets the requirements of industrial applications;the proposed method of removing outliers with multiple statistical indicators increases the accuracy rate by 3.6% and eliminates the influence of some outliers;In a balanced situation,the fusion model increases the recall rate by 8.6%.The experimental results show that the method proposed in this paper improves the accuracy of the correlation estimation of road congestion in a large-scale urban road network to a certain extent.