Study On Key Technologies Of Spatio-Temporal Data Analysis In Intelligent Transportation System

With the rapid development of economy and society, the city scale continues to expand while transportation is under great stress. Constructing the intelligent transportation system is an effective solution to improve the level of transportation management and decision support. It can alleviate the traffic accident and congestion influence, and improve the urban environment and living quality of residents. Throughout the construction of intelligent transportation system in the world, the transportation infrastructure such as road, signal lamp and camera are developed and progressed greatly. With the development of information technologies such as sensor, wireless communication, smart device, Internet and cloud computing, massive traffic data are collected. Many applications such as real-time monitor, navigation and urban emergency management are emerging, which makes a new opportunity and challenge for intelligent transportation industry.There are many valuable spatio-temporal knowledge are implied in massive traffic data, such as spatio-temporal distribution, association rule, variation trend, etc. Spatio-temporal knowledge is useful for decision supporting in traffic scheduling, road planning, target tracking, etc. Facing to the requirement of dynamic analysis of traffic data and status, this dissertation studies multi-granularity and dynamic model of transportation network, and analyzes the spatio-temporal data organization and management strategy. Measure and expression methods of similarity, correlation and association are studied from different data elements such as trajectory, traffic flow and congestion status. These methods are applied for analyzing hotspot distribution, congestion trend, short-time traffic forecasting and anomaly detection.Facing to the requirements of dynamic transportation analysis, the main research works of this dissertation are listed as follows.(1) The data organization and management strategy of multi-granularity and dynamic transportation network is studied. It adapts to the requirement of knowledge acquisition for different user interests and spatio-temporal area.(Chapter2)Based on the core idea of "integrated management of multi-granularity and dynamic transportation data", the spatio-temporal model breaks through the bottleneck of traditional spatial data management and provides a solution for dynamic transportation data organization and management. It reduces the data redundancy and supports multi-granularity data access and multi-level storage. It is useful for constructing a ubiquitous system. On the other hand, this model introduces the data stream technology to support the static, continuous and hybrid query. It helps to improve the capability of real-time processing.(2) Considering the trajectory data and combining with their spatio-temporal features and semantics, the spatio-temporal similarity measure method is studied. Trajectory clustering based on spatio-temporal similarity analysis can be used for mining user moving pattern and spatio-temporal distribution.(Chapter3)This method firstly extracts characteristic points according to the road constraints, such as crossings and stops, to partition the trajectories. This step sharply reduces the data size. Then, the temporal similarity and spatial similarity are measured separately. The spatio-temporal distance of trajectories is calculated after normalization and is applied for clustering. Experiment results show that this method gains good performance. The clusters reflect the knowledge of moving pattern, hot path, etc.(3) Considering the traffic flow data and combining with their spatio-temporal features and semantics, a spatio-temporal correlation analysis method is studied. It can be used for short-term traffic flow forecasting.(Chapter4)In this method, spatio-temporal correlation coefficient is defined firstly to reveal the relationship of different traffic flow series, a quick calculation method of spatio-temporal correlation coefficient is proposed after analyzing the corresponding properties. Secondly, a spatio-temporal analysis algorithm based on spatio-temporal correlation coefficient matrix is proposed to choose the proper predictor. At last, a forecasting model is presented based on support vector machine due to the nonlinear characteristics of traffic flow. Experiment results show that this method can gain higher prediction precision.(4) Considering the congestion states data and combining with their spatio-temporal features and semantics, the spatio-temporal association analysis and expression methods are proposed. It can be used for spatio-temporal association rule mining to analyze the congestion trend and reason.(Chapter5) This method considers the spatial and temporal constraints simultaneously, which filters irrelevant data in advance and improves the efficiency of spatio-temporal association rule discovering. Based on the idea, this method analyzes the time validity and spatial relativity simultaneously during the generation of frequency item sets. It classifies the time duration of spatio-temporal data and considers the spatial relationship firstly and generates the transaction table, then performs join operation on spatial-related item sets. Experiments illuminate that the algorithm is well performed. The algorithm is applied in intelligent transportation system to analyze the trend of traffic congestion by identifying spatio-temporal association between road segments.(5) Considering the real-time moving trajectory and predefined route, the dynamic distance calculation method of spatio-temporal sequence is studied. It can be integrated in real-time detection of trajectory anomaly.(Chapter6)In many fields such as public transportation and logistics service, the trajectories of moving objects are constrained by road network and mostly predefined. Deviating from the normal trajectory might imply some problems. During the detection period, partial sequence of the real trajectory is selected dynamically based on continuous query model of stream data, and the scope of normal trajectory is adjusted correspondingly. The improved Hausdorff Distance is used to reflect the degree of deviation finally. Experiments show that, comparing with the conventional map matching method, the proposed algorithm is more efficient for anomaly detection.The main contribution of this dissertation is the research and application of spatio-temporal data analysis methods with the domain constrains. The research results of key techniques such as spatio-temporal data integration and management, spatio-temporal semantic expansion, spatio-temporal similarity measure, spatio-temporal correlation measure and spatio-temporal association rule are useful for improving the management and service quality of intelligent transportation system.