| With the rapid development of economy and science technology,both of expressways and provincial trunk roads construction and transformation in China have continued to accelerate,also the mileage and density of road networks have gradually expanded.A large amount of data has been accumulated in information collection devices such as toll collection systems,lane induction coils,and cameras along the road.In the meanwhile,as an important part of the intelligent transportation system,how to model the traffic travel model and use the big data mining technology to mine the trajectory data deeply,to achieve accurate prediction of the travel behavior of expressway users,which is important in decision on traffic management,user travel personalized service,ease traffic congestion,road planning,etc.Based on the expressway travel data,this paper presents the way to predict the user’s travel behavior.The main research contents are as follows:(1)User travel behavior pattern mining.Facing the massive travel data generated by expressway information collection devices,the big data processing technology is used to model and analyze the user travel mode.Meanwhile,this paper realized the user home location speculation and location semantic enhancement method,which enrich the attribute information of sparse locations.In addition,the travel purpose inference algorithm is designed to complete the in-depth mining of users’ travel patterns.(2)User travel times forecast.The user travel times prediction problem is regarded as the regression problem.The user representation module and the travel mode representation module are used to model the feature.A personalized spatio-temporal prediction model(STU-LSTM)based on long short-term memory network is proposed to realize the user travel times forecast.(3)User travel origin-destination(OD)forecast.The user travel OD prediction problem is regarded as a classification problem.By constructing the spatio-temporal representation module and the user travel representation module,a personalized spatiotemporal attention prediction model(STUASM)is proposed to realize the User travel origin-destination(OD)forecast.Experiments were conducted on the actual travel data of the expressway.The experimental results show that the modeling method of user travel behavior pattern mining proposed in this paper,including travel preference and travel purpose analysis,is helpful to understand the user’s travel mode;STU-LSTM model is proposed for expressway users’ single source time series prediction problem.The personalized representation learning module and the LSTM-based network design have certain advantages in mining sequence hidden time series information;the STUASM model in the user travel OD forecast,the proposed spatio-temporal representation module can fully model the spatio-temporal information of the sequence,user travel mode designed means that the module can mine the user’s personalized dynamic mode,and the attention mechanism is introduced can capture the correlation of spatio-temporal sequences,and the enhanced position semantic representation has a significant improvement on the prediction results of the improved model,and the experimental results are significantly better than other existing models. |
