A Real-Time Driving Destination Prediction Model Based on Historical Travel Patterns and Current Driving Status

Abstract

As an important application of location-based services (LBS), driving destination prediction support route planning, service recommendation and vehicle scheduling, etc. However, destination prediction in a real-time manner is challengeable due to the influence of both historical travel patterns and current driving status, which need to be considered in modeling. To fill this gap, we proposed a real-time prediction model with two modules, i.e., hierarchical temporal attention module and status-aware prediction module, which utilize driver’s historical Original-Destination (OD) sequences and current travel trajectories as inputs respectively. More specifically, the hierarchical temporal attention module can effectively process the OD sequences under the calendar period. The status-aware prediction module achieves the prediction according to the current travel status and the key travel location identification in current trajectory. Comparative experiments with baseline models verified the validity of our model. Further analyses discussed the factors that affect the prediction performance from the perspective of distance, time span and grid partition granularity.