Research On Pattern Mining And Traffic Forecasting Of Urban Rail Transit Network

With the continuous development of urbanization,cities are beginning to face various risks and challenges.Traffic congestion,traffic safety and environmental pollution have become the three most difficult problems facing most cities in China.Vigorously developing urban rail transit has become one of the important means to solve these difficult problems.Realizing the intelligent,integrated and efficient management of urban rail transit networks is also an important part of smart cities.Based on the Shanghai intelligent transportation card data and the spatial location of the rail transit network,this paper conducts statistical analysis and modeling prediction of passenger movement behavior and station passenger flow in urban rail transit network.This paper uses statistical methods,machine learning,tensor analysis,pattern recognition to discover passenger travel pattern,forecast peak traffic flow and estimate peak time in urban rail transit networks.Firstly,this paper studies the basic characteristics of data and the correlation between time and space of passenger traffic in urban rail transit network from the statistical point of view.In the morning peak,evening peak and off-peak hours,the tidal effect and the change of passenger flow are analyzed respectively,and the the popular stations are counted.Moreover,we used aggregate sampling to study two typical types of passenger travel patterns.For the passenger travel pattern mining of urban rail transit network,we first analyze and verify the historical traffic flow characteristics,extract the priori features contained in the traffic flow data,and then reconstruct the traffic flow data into 5-way tensor based on the prior characteristics.The traffic tensor incorporates the basic trend of time series contained in traffic flow data,multiperiod characteristics(week periodicity and day periodicity),spatial structural characteristics and abruptness.After that,we obtain the time mode and the spatial mode based on the tensor decomposition method.The high-order orthogonal iteration(HOOI)algorithm is used to solve the optimization problem.Then,we analyze the changes of traffic flow under the influence of various modes through experiments,and excavate the passenger's travel pattern and prove the reliability of traffic flow prior knowledge.For the peak traffic flow forecasting and peak time estimation of urban rail transit network,we propose a tensor-based framework combined with“priori modeling”and“posterior analysis”based on the prior knowledge of traffic flow verified in the passenger pattern mining work,which can forecast peak-hour passenger flow with both accuracy and interpretability.Based on the constructed 5-way tensor,we designed peak tensor completion algorithm to predict the future peak flow and estimate peak time.The Alternating Direction Method of Multipliers(ADMM)framework is used to solve the optimization problem of tensor completion.Finally,in order to further understand the behavior of proposed tensor model and evaluate the importance of priori features(modes of tensor),we propose a posterior analysis process : quantitatively estimates the importance of mode of tensor by evaluating the tensor models that remove specific mode.Finally,the two experimental schemes of prediction time advance and prediction interval length are designed to evaluate predictive performance from multiple angles.Experimental results illustrate that our approach shows superior performance both on peak-hour and off-peak hours,with forecasting error reduced by 25%-40% compared to baseline models and our model is robust in terms of timing advance and interval length.Posterior analysis process shows that proposed tensor model does combine effective multimode information,with destination relevance and week mode having a significant impact on performance.