Research On Traffic Situation Analysis And Application For Urban Road Network Through Spatio-temporal Data Mining

Urban road network traffic system is a highly complex nonlinear system.Accurately assessing the spatio-temporal traffic situation of road network and deeply mining its complex operation regularity has important theoretical significance and practical value for improving traffic operation efficiency and intelligent traffic management technology.With the rapid development of sensing,communication and intelligent computing technologies,how to explore the potential traffic evolution patterns and characteristics from massive,high-dimensional,and diverse spatio-temporal traffic data,and provide efficient and accurate information services for traffic management departments and travelers is a significant challenge in the field of intelligent transportation.This study proposes a data-driven research framework for the research on traffic situation analysis and application for urban road network through spatio-temporal data mining,applying the advanced machine learning,information theory and other data mining and analysis methods,combined with traffic flow theory and network theory.The thesis is organized according to the logic thread,as follows: spatio-temporal feature extraction,causal relationship mining,frequent pattern recognition and future state prediction.Firstly,a homogenous spatiotemporal pattern extraction method based on nonnegative tensor decomposition is proposed considering the multi-dimensional characteristics of road network traffic state.The long-term and short-term variation features are respectively extracted in temporal dimension.The spatial proximity coefficient is defined in the spatial dimension to modify the Gaussian similarity matrix.The large-scale urban road network is divided into sub-networks with homogeneous spatio-temporal traffic flow pattern based on the improved spectral clustering algorithm.Secondly,a spatio-temporal causality mining model is established for urban road network traffic flow considering the direction of information transmission.The transfer entropy algorithm is introduced to quantify the spatio-temporal causality of traffic state.The sliding window technique and Gaussian kernel density estimation method are used to calculate the transfer entropy matrix,characterizing the dynamic information transmission between any two segments.The affected coefficient,influence coefficient,input degree and output degree indexes are proposed to identify the key sections of the road network.Thirdly,a traffic congestion propagation pattern recognition model is proposed based on spatio-temporal causality mining.The transfer entropy is used to calculate the amount of traffic state information transmission and the intensity of causal correlation between traffic congestion events.Considering the constraints including the continuity of the occurrence time and the connectivity of the spatial topology for the congestion events,the redundant correlation relationship is removed based on the causality significance test,and the directed graphs of spatio-temporal traffic congestion propagation for the road network is established.The frequent subgraph mining algorithm is used to identify the frequent spatio-temporal traffic congestion propagation patterns.Finally,a traffic flow predictive model for urban road network considering spatial and temporal information is constructed.The historical upstream and downstream traffic flow parameters are selected as the input variable,and the gradient boosting decision trees(GBDT),an ensemble learning algorithm,is applied to predict the short-term traffic flow and identify the importance of variables.Then the spatiotemporal feature variables with high contribution degree are extracted by the variable selection method based on the transfer entropy,significantly reducing the feature dimension and complexity of the model,and improving the prediction accuracy and efficiency.This study provides a systematic research framework and methods for spatiotemporal traffic data mining and analysis,and promotes the fusion of traffic situation modeling and spatio-temporal data mining,providing the scientific decision support for urban traffic congestion control and road network planning,and the basic technology support for data-driven intelligent transportation systems.