Research On Private Car User Travel Behavior Prediction Based On Semantic Perception

With the development of urbanization and rural revitalization,the increasing number of private cars has brought tremendous pressure to the existing traffic system and caused severe problems such as traffic jams and environmental pollution.Under the background of the construction of the intelligent transportation system,the popularization of vehicle-mounted terminal trajectory data to collect a large number of private cars.Thus,the application of data mining technology analyze the travel behavior,the user is looking for laws of private cars in urban traffic system,by leading groups in the city of private cars flow distribution to solve the traffic problems become possible.Existing research on travel behavior based on private car trajectory data mostly focuses on the analysis of spatiotemporal information,and usually ignores the semantic information of location.Therefore,it is difficult for the model to perceive users’ travel rules from the semantic perspective in the process of modeling user travel behavior.If the travel rules of users are understood from the subjective perspective of users based on semantic information,the travel intentions of users can be grasped from the root,and the research results that are more close to human subjective thinking can be obtained.Therefore,starting from the analysis of users’ travel intentions,this paper focuses on the learning degree of the model on users’ travel patterns under semantic perception scenarios,to obtain more accurate travel behavior prediction results.The main work is as follows.In view of the "drift phenomenon" existing in the stopover points of private cars,that is,the same semantic location corresponds to multiple stopover points of private cars,resulting in the difficult matching of semantic information of stopover points,this paper designed a semantic perception method based on DBSCAN clustering principle and AOI(Area of Interest)boundary information.On this basis,the spatial semantic information of each location is defined,and the visual result of semantic perception is given.To solve the problem of how to integrate semantic information in the process of private car trajectory sequence modeling,this paper proposes a Semantic Long Short Memory Network(Sem-LSTM)model based on Semantic perception.In this model,semantic information is represented as a vector and embedded into the model.By using dynamic hidden layer variables and steady-state user preference variables of Sem-LSTM network layer to model user travel behavior globally,the model can capture semantic information in the training process,and finally improve the prediction accuracy of user travel behavior.This paper evaluates the performance of Sem-LSTM model in predicting user travel behavior and compares it with other six related methods to demonstrate the performance of Sem-LSTM in all aspects.The experiment proves that the Sem-LSTM model can learn the sequence characteristics and spatial-temporal characteristics of the trajectory data and capture the travel rules of users from the semantic perspective.Therefore,compared with the six methods,the Sem-LSTM prediction results is excellent and stable.