Fastest Route Search Algorithm Based On Uncertain Traffic Condition Modeling

In recent years,with the rapid progress of urbanization,traffic congestion has become increasingly serious,which not only wastes traffic resources,but also seriously affects people’s commuting efficiency.In this context,the fastest route search,as an important part of the intelligent transportation system,aims to analyze the historical patterns and real-time status of urban traffic flow for the fastest route in complicated road network.Meanwhile,the highly dynamic traffic flow in the city may have an unavoidable impact on path selection.Therefore,while searching for the fastest path in the real-time road network,it is also necessary to consider the uncertainty of future road conditions,avoid potential congested road sections to ensure the traffic efficiency of the output route.Based on the above background,this paper mainly studies the fastest route search in real-world scenarios and how to provide future traffic information for route search.In the fastest route search problem,existing methods try to model the historical patterns of traffic flow using deep networks and combine them with pathfinding algorithms to search for the fastest path.However,these methods cannot effectively model the short-term trends and long-term patterns in traffic conditions.This paper proposes a spatial-temporal gating network to dynamically perceive the short-term and long-term patterns of traffic flows by upperlevel tasks to control the transmission of traffic information.In addition,existing methods cannot meet the upper-bound constraint of the pathfinding algorithm,resulting in suboptimal output paths.To address this,a probability-based method is proposed for the fastest route search,which models the distribution of travel time using variational autoencoder and obtains constrained values from the distribution.Considering the importance of future road conditions for the path search,this paper further studies the problem of traffic prediction.Existing traffic prediction methods ignore the structural features of different turning lanes at intersections,this paper’s model incorporates the structural information of intersections into the modeling of urban road network,allowing it to perceive the structural features of different intersections.For the fastest route search and traffic forecasting problems studied in this paper,this paper respectively verifies the effectiveness of the proposed models on two real-world datasets.The relevant experiments compare our model with the most important existing methods.The results indicate that the proposed traffic prediction model achieves higher prediction accuracy than the existing models;and the proposed route search models are superior to the previous methods in terms of search efficiency and path optimality.Therefore,this paper has certain value for real-world applications.