Research On Travel Mode Recognition Based On Convolutional Neural Network

With the rapid development of the mobile Internet,mobile terminal devices with global positioning system(GPS)positioning functions are ubiquitous,which provides more convenience for studying traffic problems.User travel mode is an important part of traffic demand analysis and transportation planning,so the study of user travel mode is of great significance.With the rapid development of urban transportation,the data of residents’travel mode,travel times and travel distance become extremely huge.Traditional methods of investigating residents’travel data are mainly based on manual work,but in the face of the rapid growth of residents’travel data,problems such as insufficient objectivity,high cost and long survey cycle of traditional methods are gradually exposed.In recent years,the development of GPS positioning technology has promoted the reduction of the cost of GPS equipment and services,and the popularization of smart phones with positioning function,which has further promoted the research of intelligent identification and automatic data acquisition of residents’travel modes.The data recorded by GPS equipment has the advantages of high quality,fast update and wide coverage,which makes GPS trajectory recognition become a new method to solve the traditional travel survey problems.GPS data records user’s space-time trajectory data.How to automatically identify the semantic information of user’s travel mode from GPS trajectory data has become the focus and difficulty of research.The most important difficulties in the study of travel mode identification are parameter selection and time series analysis.The inappropriate selection of parameters and incorrect time series analysis methods will hinder the improvement of the accuracy of travel mode recognition.In view of the above difficulties,this thesis designs a deep convolution neural network model to recognize the mode of travel from GPS trajectory,and realizes the recognition of bicycles,walks,cars and buses.In Firstly,this thesis preprocesses the original GPS trajectory data,including threshold processing and smoothing processing,eliminating the abnormal data and the data which are greatly disturbed by the outside world,and the GPS trajectory data with higher quality is obtained,which lays the data foundation for the research.Secondly,the data after pretreatment are analyzed in depth.According to the analysis results,the four characteristics of speed,acceleration,urgency and steering angle of different travel modes are compared,and the reason why these four characteristics can be used to distinguish different travel modes is elaborated.Thirdly,on the basis of the above analysis,several groups of convolutional neural networks with different levels are selected,and the four parameters mentioned above are used as the input layer of each group of convolutional neural networks.By comparing the recognition results of each group of convolutional neural networks,the best group is selected.Finally,the convolution neural network model is optimized,and the convolution neural network model with the best recognition effect is obtained.By comparing the results of the convolutional neural network model with those of traditional travel mode recognition and other research results,it is found that the convolutional neural network model in this thesis has better recognition effect,and also proves the superiority of the convolutional neural network model in the field of travel mode recognition.