Dynamic Ridesharing Recommendation Method For Commuting Private Vehicles Based On Cross-domain Urban Data Fusion

Urban infrastructure construction greatly facilitates daily transportation for residents.Meanwhile,it causes the problems of road congestion,gas emission and air pollution.Thus,it is important to fuse and use the huge amount of traffic data generated by urban roads to provide feasible decision support for alleviating traffic congestion and reducing running vehicles,with big data analytics technology.As the rapid development of location-based services and mobile internet technology,ridesharing service is considered as an effective approach to cope with the traffic congestion in the sharing economy.Nevertheless,there are two challenges in existing ridesharing services.Firstly,existing approaches mainly employed an on-demand trip matching algorithm to satisfy individual passengers with short waiting time.However,they paid inadequate attentions on environmental concerns that the provisioning of on-demand ridesharing relies much on a large number of private vehicles,which might aggravate the sufficiently congested urban traffic network.Secondly,the data sources for existing ridesharing service rely much on the large-scale human mobility data such as GPS trajectories,and CDRs,which are trajectory-abundant,but semantic missing.They failed to effectively consider the impact of weather,anomalies,traffic conditions and accidents on the ridesharing.Thus,they could not provide personalized and dynamic rideshare services for passengers.To address the problem mentioned above,in this paper,we focus on facilitating the dynamic ridesharing among commuting private vehicles during peak hours in working days.Fusing cross-domain urban data(e.g.weather,accident and traffic condition),we propose CommuteShare,a dynamic ridesharing recommendation algorithm for commuting private vehicles.CommuteShare aims to propose a long-term ridesharing mechanism,which could not only enhance passenger's satisfaction(i.e.short waiting time and matching accuracy),but also guarantee comprehensive social benefits(i.e.reduce tons of non-shared private vehicles during peak hours in working days).Our main research contents contain:Firstly,we propose a commuting private vehicles identification algorithm,using large-scale vehicle license plate recognition(VLPR)data.Secondly,we propose a dynamic ridesharing recommendation algorithm for commuting private vehicles,using spatio-temporal features fused with dynamic influence factors(e.g.weather,accident and traffic condition).Thirdly,we compare the proposed CommuteShare with four other baseline methods(k-means based on Squared Euclidean Distance,k-means based on City Block Distance,k-means++ and AGNES)Experimental results show that the proposed method significantly outperforms existing approaches in both enhancing rideshare accuracy(i.e.,83%)and reducing waiting time(i.e.,7 minutes).Furthermore,our method could reduce 30%commuting private vehicles average daily in the morning peak hours.