Research On The Method Of Travel Mode Choice Based On Mobile Phone Signaling Data For Online Traffic Simulation

The acquisition of real-time OD has always been the bottleneck of the developme nt of domestic online traffic simulation technology.At present,the macro models,the real-time generation models and the fixed detectors are not preferred solutions to obtai n real-time OD.In recent years,with the continuous development of traffic informatio n collection technology,the technology of extracting travel OD through mobile phone signaling data has been relatively mature.It has the advantages of nearly full sample,good spatiotemporal characteristics and real-time acquisition,etc.Which can be consid ered as a good input data for online traffic simulation,on the premise of solving the problem of mobile phone signaling data in travel mode choice.In order to solve the problem that OD extracted from mobile signaling data cann ot be divided into traffic modes.This paper combines mobile phone signaling data and network crawling data to extract the travel attributes and personal attributes of mobil e phone users,as the input data of the online simulation travel mode choice model.B ased on the good prediction and generalization ability of the machine learning model,three travel mode choice models are built with the resident survey data,and evaluated by multiple indicators.The main contents of this article are as follows:Firstly,the applicability of mobile phone signaling data is discussed according to the principle of mobile communication.And the superiority of the machine learning m ethod is analyzed by comparing the traditional travel mode choice model.Therefore,th e preliminary feasibility and key points of travel mode choice based on mobile phone signaling data and machine learning methods are demonstrated.Secondly,in the feature selection,the importance analysis of the existing feature data and the easy accessibility analysis of the model application feature data are carrie d out to preserve the important easy-to-acquire features and replace the important diffi cult-to-obtain features as the principle to ensure the existence of important feature valu es;In the data processing,the difference between the model construction data set andthe model application data set is eliminated from the data set parameters and the wh ole to ensure the equivalence of the model construction data set and the model applic ation data set,thereby ensuring the validity of the model result.Thirdly,analyze the impact of the parameters of each machine learning model on accuracy.Using grid search algorithm for parameter optimization to construct BP neu ral network,support vector machine,random forest model,and use indicators such as accuracy,precision,recall,and learning curve to evaluate the model in many aspects.Finally,some representative traffic zones in Chengdu are selected to verify the m odel results,so as to analyze the applicability of the results on the spatial and tempor al scales,and based on the model application requirements and framework to analyze the applicability of the model.The results show that the prediction results of the trave l mode choice model are stable and reliable,and have certain application value.