Research On The Traffic Trip Mode Identification Based On Mobile Phone Data

Optimizing the structure of residents trip modes based on the individual travel demand is the key to solving the traffic congestion problem.The mobile phone signaling data is a kind of positioning data covering a wide range of people,with the advantage of reliable and real-time.However,the data is still limited to medium and macro decision analysis in the field of urban and traffic planning,which is still less involved in the extraction of refined travel information,especially resident trip mode.This paper aims to establish a trip mode identification model based on 4G mobile phone signaling data.Focusing on this core goal,this paper conduct the research from travel track extraction based on mobile phone data and trip mode recognition mod el construction.The main research contents and conclusions are as follows:1)Identifying the parking point from the original positioning point,extracting the travel information and obtaining the user travel trajectory on the road network.Firstly,data preprocessing algorithms are designed for different types of noise data in the original data.Secondly,according to the spatio-temporal characteristics of the points and the definition of travel,spatio-temporal clustering algorithm is designed to classify the track points to obtain the user travel information.Finally,matching the base station and the intersection to obtain the real travel trajectory of the mobile phone user on the road network based on the coverage of road frequency and spatial distance.2)By integrating the multi-source data to obtain the training samples and extract the feature parameters,then constructing the trip mode recognition model based on the machine learning algorithm.Firstly,5861 samples with trip mode labels are obtained by matching the resident travel survey data.Secondly,on the basis of the merge of GIS information and navigation information,29 feature parameters are extracted and the genetic algorithm is used to select the features with 16 characteristic parameters retained.Then,KNN algorithm,support vector machine algorithm and random forest algorithm are used to construct the trip mode model.The overall recognition precisions of the three models are 72.67%,68.66% and 80.79%.The accuracy of walking is the highest,the bus recognition effect is poor,and the precisions of cars and electric vehicles are about 80%.Finally,the importance of the feature parameters in the random forest model is ranked.The results show that spatio-temporal characteristics of the travel track are the most important,followed by the path navigation characteristics,and travel behavior characteristics of residents in the last.