Research On Urban Traffic Congestion Based On GPS Data Of Online Ride-hailing

The rapid growth of the number of motor vehicles and the restriction of urban road construction have led to serious contradiction between supply and demand of roads,which has led to the frequent occurrence of congestion.How to use accurate historical traffic flow parameter data to accurately and quickly identify and predict traffic congestion and the timely use of reasonable traffic interventions through the identification and prediction of results is the key to alleviating traffic congestion.At present,the booming development of the network car has solved the problem of people's difficulty in taking a taxi to a certain extent.On the other hand,the car-hailing service has generated a large number of GPS positioning point travel data,which has a large amount of data,a wide coverage and a high accuracy.So it is very suitable for the study of traffic congestion.This thesis is based on the GPS positioning point data provided by the largest network share Didi Dache of the car-hailing service market.By analyzing the characteristics of the data and fully mining the information contained in the data,combined with advanced machine learning algorithms,GPS data preprocessing,traffic flow parameter extraction,traffic congestion state division and identification,and traffic flow parameter prediction were studied.This thesis first analyzes the problems of timing chaos,data redundancy,data drift and other problems in the GPS data of the car-hailing service.For solving these problems some corresponding processing methods are carried,and then the coordinate conversion,map matching and other work are carried out.Next,this thesis extracts four traffic flow parameters of flow,velocity,density and time delay index of a single road segment in the clean GPS data.For solving the problems existing in the extracted parameters,blank time filling and wavelet denoising are further processed.After the above processing,this thesis proposes a traffic state change recognition model based on Bagging integration.The Bagging integrated model randomly samples 70% of the original traffic state change data set by bootstrap sampling,and further randomizes feature subspace of the data set by introducing random subspace method.and the kNN classifier algorithm is used as the base classifier.In this thesis,the data clustering state change datasets used in the model are constructed by using spectral clustering and ADASYN oversampling algorithms.By comparing the performance of the model in this thesis with that of other models in this data set,it can be found that the model in this thesis performs better in the identification of traffic state change.In addition,this thesis proposes a multi-input and multi-output traffic flow parameterprediction model based on GRU module.In this thesis,it is expected to build a multi-input and multi-output traffic flow parameter prediction model by analyzing the prediction demand.Since the neural network can well carry out multi-input and multi-output prediction,after considering the characteristic of GRU network,this thesis uses GRU neural network as the core module to build a three-layer traffic flow parameter prediction model with input layer,hidden layer and output layer.This model forecasts and outputs the four-dimensional traffic flow parameters in the current time period t with the multi-dimensional characteristics of traffic flow parameters in the first k time periods in time period t.By comparing with other models,it is found that the model constructed in this thesis predicts better.