Research Of Massive Data Analysis Methods In Internet Of Vehicles

Internet of Vehicles(IoV)uses technologies of many fields,such as intelligent transportation,mobile Internet,and Internet of Things to monitor the information of vehicles and the traffic in order to achieve information exchange among vehicles,roads and the people,which greatly improves the efficiency of traffic and transport safety.However,there is a lot of data in the interaction among vehicles,roads and the people,so massive data analysis must be used in IoV in order to filter invalid data and extract valuable information.Data analysis process contains anomaly detection,correlation mining,etc.Due to the complex working environment and instability of IoV,there is a large number of abnormal data in the collected data.So this paper will focus on the study of abnormal data detection in IoV.Firstly,this paper studies the traffic data collection technology of IoV.Among these data collection technologies of IoV,the data collection technology based on GPS is widely used in IoV,because of the advantages of real-time and all-weather data collecting.However,the data collection technology based on GPS requires the user to load the GPS device in the vehicle,which results in a large equipment investment and maintenance costs.With the development and popularization of smart devices,the vast majority of smart phones are equipped with a variety of sensors.In this paper,we designs a data collection method by using the Android system,mobile sensors and Baidu map SDK to achieve the functions of real-time traffic data collection and data upload at a lower cost.Secondly for data quality issues of Io V.the traditional abnormal data detection algorithms require data-sample to meet the premise that its distribution obeys normal distribution.In real life,due to the randomness driving behavior and vehicle performance,data in IoV doesn't necessarily obey normal distribution.In this scenario,if still using the traditional abnormal data detection algorithms,the detection effect is poor.Kernel density estimation can be estimated probability density directly from the data-sample without reliance on any assumptions about data distribution.Therefore this paper put forward anomaly detection algorithm based on kernel density estimation.However,there is boundary effect in kernel density estimation in the actual simulation process.In this paper,the boundary effect of kernel density estimation has been improved so that it can estimate the data probability density distribution effectively within the specified interval.In the meantime,the improved kernel density estimation algorithm is used to detect abnormal data in IoV.Finally,combining with the actual collected traffic data,the simulation of the performance of this algorithm has carried on by Matlab,such as detection rate and false detecting rate.The simulation results show that this algorithm can solve the problem that the abnormal data detection algorithm based on Pauta criterion has higher false detecting rate and unstable detection rate.In other words,this algorithm has better detection effect.