Traffic Mode Recognition Based On GPS Trajectory Data

With the rapid development of the urbanization stage,the expansion of urban land has attracted a large number of people and promoted the scale of construction of transportation facilities.Urban transportation has shown the characteristics of rapid growth in travel demand.However,the rapid development and the ever-increasing traffic demand are prone to imbalances in supply and demand,which can cause urban traffic diseases such as traffic congestion and environmental pollution.Therefore,in the issue of traffic management,people need to pay more attention to the coordination within the transportation system,while seizing the new opportunities of urban transportation development in the information age,and using information technology to analyze and tap the characteristics of residents’ travel.The identification of transportation modes is part of the research content of residents’ travel.In the classification based on GPS trajectory data,the commonly used feature types include the average velocity,instantaneous velocity,acceleration,distance and time of the trajectory segment,which are all calculated from the space-time information of GPS collection points.The description of the trajectory of this part of the feature can be classified as the time domain dimension.The key point is that expanding the dimension of features and designing models for feature fusion is the focus of this article.Therefore,starting from the basic features generated in the time domain,this thesis proposes a frequency domain feature and a multi-dimensional feature fusion model,so as to integrate the time domain and frequency domain features and classify their transportation modes.The main work can be summarized as the following four points:(1)Perform data cleaning and data expansion on the original data set to improve the quality and scale of training samples.(2)Using short-time Fourier transform and overlaying sliding window to smooth the features,so as to obtain the frequency domain features with high time resolution and balanced frequency resolution.In addition,the long short-term memory network is used to filter the frequency domain features generated under different frequencies,thus confirming the high and medium amplitude frequency domain features as the final expression of the frequency domain.(3)According to the analysis of different traffic modes and different characteristics,a combined neural network model based on multi-scale convolutional neural network and Transformer encoder framework is proposed.The time-domain features and frequency-domain features are spliced into the model,and their fusion features are deeply extracted and compressed at different scales through the function of convolutional neural network.The long-distance features are further extracted in the framework of Transformer encoder,and then the classifier is designed to realize the classification and recognition of different traffic modes.Then the classifier is designed to realize the classification of different traffic modes.(4)To verify the effectiveness of this model in experiments,and at the same time to increase the contrast of the model,the traditional machine learning model and the deep learning model proposed in previous studies are used for comparative analysis with the model in this thesis.The experimental results show that the combined neural network’s recognition accuracy rate for different transportation modes has reached 90%,which verifies that the model itself has good classification and recognition performance.The comparison results with the traditional model show that the recognition accuracy of the combined neural network model is significantly better than that of the traditional classification model.In the case of considering the amount of data and training speed,the model in this thesis is more suitable than the traditional machine learning.Compared with the deep learning model used in previous studies,this model has the advantages of large amount of training data and multi-dimensional feature extraction.In conclusion,the proposed combined neural network model shows good classification and recognition performance in the experiment,which can provide support for residents ’ travel research and urban traffic management optimization.