Research On Massive Data Real-time Processing In Internet Of Vehicles

As an important branch of the Internet of Things,the research of Internet of Vehicle(IoV)has been paid more and more attention.For enterprise,the IoV system need to collect and distribute large amounts of real-time information,the system scale determined by concurrent access quantity and message processing requirements of the vehicle terminal.Processing the information received from IoV,has a wide range of applications,such as real-time monitoring and analysis of driving behavior,alarm reminder,etc.Some of these data require real-time analysis,such as real-time monitoring and alarm reminder.With the high frequency of real-time data processing,traditional methods have high delay,slow response and other deficiencies,which cannot meet the needs of the performance of the IoV system.In this dissertation,we research the massive real-time data processing problems in IoV system based on a real project undertaken by our research center.It mainly includes the massive real-time data visualized in the map,the processing speed of the vehicle area limited alarm processing and the time delay of the vehicle service station alarm processing.1.The user can zoom or drag the map to monitor his vehicle distribution in different field.The aggregation function provided by Baidu maps has long response time,which cannot meet the real-time monitoring performance of IoV system requirements.This dissertation improved k-means aggregation algorithm based on the administrative region partition,which may produce some overlap on map,but it can accurately reflect the original distribution of the vehicle and the response is rapid and timely.2.When the vehicle is out of the restricted area,the IoV system should be alert in time.In the pre-processing,the polygon region is converted to grid storage by using the scan conversion algorithm,and each trellis is encoded by Morton code,then the quadtree compression algorithm is used to compress the storage.When the vehicle position data is received,determine the vehicle is inside or outside the area by quadtree searching.Compared with ray-crossing method,our improved method can reduce the decision time from O(N)to O(log4N).3.When the vehicle enters the radius of the service station,IoV system should mark the car.According to the location of the vehicle to query the nearest service station,this is a Top-k search problem.Vehicle data is returned every thirty seconds,so the query is very frequent,traversing all service stations is time-consuming,result in low efficiency.In this dissertation,we propose a data divide based on k-means,set up the KD tree for the data of the cluster centers and the data of each cluster,and then use the KD tree to find the station instead of the Euclidean distance calculation to improve the query efficiency.The three improved method we proposed is feasible,and has been applied in practical IoV system,the system based on OpenStack cloud computing platform,ActiveMQ message server,MySQL and MongoDB database,Redis cache.The system has the ability to support access to 100 thousand cars,and can support the rapid expansion of more vehicle access through horizontal expansion.