Research On Real-time Carpooling Technology Based On Road Network Traffic Situation Awareness

The increasing number of vehicles makes traffic big data gradually attract people’s attention,and it is also more likely to cause traffic congestion.Carpooling is a shared travel mode that combines the flexibility and efficiency of private cars and the low cost of shared transportation at the same time,making it the best choice for more people to travel.Most of the existing research on carpooling problems cannot cope with the changing traffic conditions such as traffic jams and traffic accidents in the actual road network environment,and the flexibility is poor.In response to these problems,a real-time carpooling technology based on road network traffic situation awareness is proposed,which is divided into offline stages and online stages.The offline phase trains the prediction model of roadnetwork traffic speed,and the online phase realizes the functions of filtering of candidate driver sets,real-time matching of driver-passenger pairs,and route planning based on situational awareness.The main work of this real-time carpooling technology is as follows:(1)In order to improve the real-time carpooling system’s ability to handle large-scale concurrent requests,the carpooling system is divided into offline and online stages.The offline stage mainly conducts the training of the speed prediction model,provides a reference for situational awareness in the online stage,and saves the time spent on matching in the online stage.(2)In the offline stage,in view of the problem of poor spatial applicability of existing traffic flow prediction methods and easy data leakage in a centralized environment,a traffic speed prediction method based on federated learning is proposed: first,classify road segments according to the topology to retain the traffic mode characteristics of similar road segments,and then use the particle swarm algorithm to select the mobile terminal set to improve the model quality,and finally use the LSTM(Long Short Term Memory)model for training to obtain the speed prediction model.(3)In the online stage,in view of the problem that the existing carpool algorithms cannot cope with real-time road conditions,the concept of shared trip ratio and a real-time carpool algorithm based on situational awareness are proposed.First,a pruning strategy based on the European distance ratio of shared trips is proposed to achieve efficient filtering of candidate drivers.Then,the situational awareness algorithm is called for the set of selected drivers to predict the road conditions of the initial path obtained by the pre-matching.Finally,the best driver and corresponding driving route matched for each passenger.Compared with other existing carpooling algorithms,this real-time carpooling technology has the following main contributions: 1)It adds a situational awareness link,which can effectively adapt to complex and changeable road conditions and reduce the user’s travel time cost;2)The particle swarm algorithm is used to improve the federated learning framework,thereby improving the accuracy of the prediction model;3)A pruning strategy based on the Euclidean distance ratio of shared trips is proposed,which can efficiently select candidate drivers and improve the matching efficiency.