Research On Application Of Real-Time Data Warehouse System For Expressway Vehicle Location Prediction

Currently,vehicle position prediction has become an important part of intelligent transportation.In China’s 14 th Five-Year Plan,the development direction of real-time data exchange based on vehicle-to-vehicle and vehicle-to-road integration was clearly defined.The ETC gantry system,as one of the new key infrastructure deployments on highways in China,has been widely deployed on national highways,enabling fullspectrum perception of all vehicles on the highway,which has laid a good foundation for research and application of on-road vehicle position prediction on highways.However,the current on-road vehicle position prediction based on the ETC system is still incomplete,especially with problems such as a single data processing method,low flexibility,and high time delay,which makes it difficult to meet the efficient processing needs of on-road vehicle position prediction on highways.Therefore,this paper focuses on the ETC system and its transaction big data on a certain province’s highway,aiming to conduct in-depth research on the real-time data warehouse system application for onroad vehicle position prediction on highways.Firstly,this paper designs and studies a real-time data warehouse system for onroad vehicle position prediction on highways.The system mainly consists of four big data components,Flink,Kafka,Hbase,and Clickhouse,and is divided into four system modules: data collection,data storage,data processing,and data application,providing environment assurance and real-time support for on-road vehicle position prediction.The data collection module collects diversified ETC data in different ways and stores the raw data in My SQL;the data storage module uses Kafka to store ODS layer data,Hbase to store dimension data,and Clickhouse to store DWS layer data;the data processing module realizes the dimensional modeling research of ETC data,based on which the ETC section information is extracted,and each layer of data is calculated and transmitted using Flink and Kafka;the data application layer conducts research on onroad vehicle position prediction based on the data storage module,including sectionunaware prediction research and section speed prediction research.Secondly,in the ETC field,due to the complexity and dynamics of sections,it is often difficult to obtain real-time section speed.To address this problem,this paper designs a flow flag clustering algorithm to cluster vehicle speed values in sections in real-time,calculates the real-time average speed of sections,and ensures an accuracy of 95.68%.At the same time,by analyzing the correlation of section features,a multiple linear regression model is built to summarize the feature correlation between sections,and the real-time speed of the next section is predicted by combining the section’s nonsensing speed result with other features,achieving an accuracy of 98.71% through comparison with real data.Finally,based on the real-time data warehouse environment,the vehicle position is predicted in real-time based on the predicted section speed mentioned above.By using the problem partitioning idea of dynamic programming to effectively divide the vehicle driving sections and taking the section speed as the vehicle driving speed,the iterative update of the vehicle state is achieved,and the second-level prediction of the vehicle position within the section is completed,with an accuracy rate of 94.34%,which has the advantage of low error and low time delay compared to other models.