Vehicle Destination Prediction Based On Spatiotemporal Semantics And Important Travel Features

With the advancement of data collection technology,more and more personal movement data are collected and recorded,which provides a data base for the study of individual travel behavior patterns.As an important branch of this research field,individual travel destination prediction provides decision support for Location-based Services(LBS),such as personalized service recommendations and traffic navigation.However,accurate destination prediction in real time is very challenging.On the one hand,the individual movement intention is deeply influenced by the multiple spatiotemporal context.On the other hand,due to the spatial structure constraints of road network and the periodic movement law of individual,individual travel has complex spatiotemporal patterns.The spatiotemporal semantics of the trajectory and the important travel spatiotemporal features can depict and affect the individual travel patterns in depth,but the research has not fully considered the important role of these factors,as a result,the modeling granularity and accuracy of individual travel process are not enough,it is difficult to meet the demand of real-time and accurate destination prediction.In view of this,this paper proposes a vehicle destination prediction method which combines spatiotemporal semantics and important travel features,and achieves end-to-end fine-grained travel destination prediction.The method includes trajectory spatiotemporal semantics extraction,important travel features detection,and model construction in three parts:1)Trajectory spatiotemporal semantics extraction: based on Points of Interest(POIs)dataset and the Frequency Term-Inverse Document Frequency(TF-IDF),constructing the location semantics map covering the research area.Based on Word2 Vec embedding method,the extracted trajectory location semantics are further encoded.The multi-level departure time semantics of day,week and month is extracted,and the heuristic circular fuzzy time encoding is carried out to realize the learning of time continuity and anisotropy of distribution of users’ temporal feature.2)Important travel features detection: extracting the deeper driving status information from the original travel trajectory data,including turning angle,driving speed and traveled distance.Based on the driving status,the location importance of each trajectory point is calculated,which is used as the heuristic term of attention mechanism to detect and capture the important spatiotemporal features that can describe the travel process of a user,that is,the trajectory point features located in important road network positions and important time-frequency features.3)Model construction: coupling with the vehicle trajectory and trajectory spatiotemporal semantics information as the original input,using the long-term dependence of the input sequence of double-layer Long Short-Term Memory(LSTM),the important travel features are detected and expressed based on spatiotemporal attention mechanism,and the destination is predicted through the residual network.In order to verify the effectiveness of the method,the private car trajectory dataset of eight users in 2019 was studied and the prediction effect of the five baseline models is compared.The experimental results show that the method can improve the accuracy of destination prediction significantly under the trajectory dataset of different travel patterns.The ablation experiment of the model further demonstrates the effectiveness of the spatiotemporal attention mechanism and the extracted trajectory spatiotemporal semantics.This paper further verifies the rationality of the encoding methods for spatiotemporal semantics,reveals the working principle of attention mechanism,and discusses some relevant factors that affect the model prediction.Through the prototype system implementation and the field deployment of the algorithm,the potential value of this method is demonstrated.At the same time,the trajectory spatiotemporal semantics and spatiotemporal attention mechanism of this paper can provide a reference for the learning and construction of important travel features in individual mobility research.