Research On User Commuting Experience Improvement Algorithm Based On Private Car Trajectory Data

With the rapid growth of economic development,the urbanization’s process is also advancing,and the number of motor vehicles is increasing rapidly,which has affected the traffic,environment and other aspects of the city.From 2016 to 2021,the market share of private cars has always accounted for more than 70% of civil cars,and kept an average annual growth of 18.48 million,with a huge increment.On the one hand,the high guaranteed quantity of private cars has brought convenience to residents’ lives and contributed to the development of the city.On the other hand,due to the mismatch between the existing road resources and the high guaranteed quantity of private cars,the traffic situation is tense,which makes the commuting experience of private car users unsatisfactory.With the vigorous development of mobile positioning technology and big data,large-scale trajectory data generated by human activities are being collected and uploaded to the cloud.The track data contains a lot of semantic information related to users.Learning from it can help improve the commute experience of private car users.However,because the trajectory data presents an irregular structure,it is difficult to spatially model it,and the commute experience of private cars is related to many factors.Most of the existing studies are based on the regular spatial structure to learn and model the single feature of trajectory data,it is difficult to completely extract the spatiotemporal features in trajectory data.Based on the data of commuter car,this thesis improves the algorithm of commuter car based on depth learning.The main work includes:1)Aiming at the characteristic that the commuting experience of private car users is closely related to the time cost spent by private car users on commuting,this study proposes a deep learning algorithm based on multi head attention LSTM.It can predict the commuting time of the user,obtain the commuting time of the private car in the future time period,and give suggestions on the user’s future commute,so as to improve the user’s commuting experience.Specifically,the distribution of hot spots in the city is obtained by clustering the trajectory data.Then,the temporal and spatial features of multiple private cars in the hot area are extracted through the multi head attention LSTM algorithm.Based on the impact of each private car on other private cars in the same hot area,the multi head attention mechanism is used to learn and model the impact of these private cars on each other at the same time,so as to better predict the commuting time of private cars.Experimental results show that this method is superior to the existing baseline algorithm in three evaluation indexes: root mean square error RMSE,average absolute error MAE and average absolute percentage error MAPE.2)In addition to the cost of commuting time,the study also found that the commuting experience of private car users is also affected by the commuting departure time of private car users.The commuting departure time is closely related to the commuting time cost.On this basis,this thesis proposes a new model of deep improvement commute experience,which can accurately predict the commuting time and time cost of private car users,so as to provide users with commuting suggestions to improve the commuting experience of private car users.In this model,this propose a multi task learning GCN algorithm,which captures the highly complex characteristics and dependencies between commuting departure time and time cost,and then models and predicts it.The experimental results show that compared with the existing methods,the model proposed in this thesis has better prediction performance,and can accurately predict the commuting departure time and time cost of private car users,so as to effectively improve the commuting experience of private car users.