Research On Road Network Speed Estimation And Prediction Methods Based On Online Car-Hailing Trajectory Data

As traffic congestion is becoming more and more serious and the concurrent problems of congestion affect the social and economic development and people’s normal life,the Intelligent Transportation System(ITS)is gradually developed and widely used.The traffic data collection and prediction of road network is the foundation of ITS,which provides support for applications such as traffic management,traffic control,and dynamic guidance.In the methods of traffic data collection,mobile detectors are widely used due to the low cost and high coverage rate.Online car-hailing can be regarded as a kind of mobile detector and the development of online car-hailing provides new ways for road traffic information collection and traffic state estimation and prediction.Therefore,this study uses advanced algorithms to estimate and predict road traffic speed based on online car-hailing data,aiming to provide a new way for road traffic speed acquisition and theoretical reference for related research.Firstly,an online car-hailing trajectory data processing process is established by using theories and technologies such as data mining and geospatial information processing.According to the research time range and spatial region,the trajectory data are spatial-temporal mapped to obtain the final experimental data set.Based on the theory of mathematical statistics,the road network speed is extracted from the experimental data set,which provides data support for the research on road network speed estimation and prediction algorithms.Secondly,the tensor basis,tensor decomposition,and other related theories are elaborated in detail.Through the spatial-temporal correlation analysis,the tensor model of road network speed data is constructed to fully explore the spatial-temporal characteristics of the data,providing a theoretical basis for follow-up research.Then,since spatial-temporal distribution of online car-hailing is unbalanced,the problem of speed estimation is transformed into the problem of missing value imputation.Two tensor-based imputation models(CP＿WOPT and Tucker＿ALS)are introduced and applied to road speed estimation.Aiming at the low accuracy of existing models,an Tucker decomposition-based improved imputation method(TDII)is proposed.In order to improve the accuracy and stability of the model,the momentum gradient descent method is adopted,and the objective function is improved.Besides,the adaptive rank algorithm is proposed to solve the problem of tensor rank selection.Based on the experimental data,the applicability and effectiveness of the model are verified from three aspects: comparison of different tensor modes,comparison of different models,and parameter analysis.Finally,in order to avoid the non-convex optimization problem and improve the prediction accuracy,a Bayesian probability CP decomposition-based single-step prediction model(BPCP＿SP)is proposed based on Bayesian theory and probabilistic reasoning theory for the single-step prediction of road network speed.In addition,in order to make the model more suitable for practical application,the dynamic tensor model is introduced to develop BPCP＿SP to a dynamic Bayesian probability CP decomposition-based multi-step prediction model(DBPCP＿MP)for the multi-step prediction of road network speed.Based on the experiments designed,this study explores the impact of the forecast time period on the prediction results and compares the prediction results of several advanced models.The experimental results confirm that the proposed algorithm can be used for both single-step and multi-step prediction of road network speed under the premise of balancing accuracy and robustness,which has practical application significance.There are 47 figures,9 tables,and 89 references in this paper.