Research And Application Of Mining Vehicle Abnormal Behabior On Expressway Based On Traffic Big Data

With the gradual maturity and improvement of the entire monitoring network on expressway,more and more information of vehicles are recorded and saved during expressway traffic,such as the situation of entering and exiting high-speed,the time and direction of the vehicle passing through section / bayonet and key nodes,etc.The number of these data generated every day is basically over one million,and they provide effective support for expressway operation management.On the other hand,with the widespread application of big data,data mining,cloud computing and other technologies,expressway managers are no longer satisfied with using these data only in the level of statistical analysis.They are eager to mine useful knowledge from these huge amounts of data in order to improve the level of expressway operation and management.In this paper,we combine the techniques of big data and data mining technology.Aiming at mining vehicle abnormal behavior in expressway traffic,this paper builds a distributed computing framework for vehicle traffic big data mining.Then models and algorithms for mining vehicle abnormal behaviors on expressway,including abnormal driving speed,abnormal entry and exit,frequent abnormalities and abnormal trajectories,are proposed and implemented on this framework.On the basis of above,this paper combines with actual projects to mine and analyze the real massive traffic flow data of a certain expressway road network in Zhejiang Province,and realize a detection approach of fake plates based on vehicle abnormal behaviors.The main research work and innovations of the paper include:1.In view of three typical expressway vehicle abnormal behaviors such as abnormal vehicle speed,abnormal entry and exit,and frequent abnormalities,this paper proposes models and algorithms for mining vehicle abnormal behaviors on distributed architecture.On the distributed computing framework Spark,Establishes a vehicle speed abnormal behavior mining model based on the interval speed judgment between adjacent trajectory points of the simulated trajectory;Establishes a vehicle entry and exit abnormal behavior mining model based on the judgment of the order of passing vehicles entering and exiting the toll station;Establishes a vehicle frequent abnormal behavior mining model based on the statistics of high frequency occurrences of long-term single-day traffic data.Complete parallel data mining by distributing massive vehicle traffic big data to various nodes.2.In view of the abnormal behavior of vehicle trajectory,this paper proposes a parallel mining algorithm for abnormal trajectories based on IBAT and PWED,which is called PPWEDIBAT(Parallel algorithm for Anomalous Trajectory Mining Based on Isolation-Based Anomalous Trajectory algorithm and Path Weighted Edit Distance).PPWEDIBAT makes several improvements on IBAT,which include using time-sharing grouping to divide the trajectory in order to take the time factor into account;using distributed technology tools such as Spark to improve the IBAT's problem of high time complexity and low efficiency in the face of massive vehicle traffic big data;reducing the false positive rate of abnormal trajectory detection by combining PWED.PWED is an algorithm based on ED(Edit Distance),which additionally considers the traffic volume during different time periods of each path node,so that the operation cost of replacement,deletion,and addition between the two trajectories changes according to the path weight.PWED can measure the similarity between two trajectories more accurately.3.Combining abnormal behavior mining based on vehicle traffic big data with practical application,this paper proposes a detection approach of fake plates based on vehicle abnormal behaviors.In order to process massive traffic data,this paper firstly builds several types of vehicle abnormal behaviors mining models associated with fake plate vehicles on a distributed architecture,and uses them to mine the real massive traffic flow data of ZheJiang expressway.Then,BP neural network algorithm is used to build a model and train mining results of several vehicle abnormal behaviors.This approach comprehensively considers a variety of abnormal behavior factors to detect fake plate vehicles.The experimental results show that,compared with the bayonet time comparison method which uses only a single method to detect fake plate vehicles,the approach in this paper has high accuracy,which effectively reduces the misjudgment rate of fake plate vehicle detection and greatly improves the work efficiency of inspectors.