Design And Simulation Of A Traffic Collecting And Predicting Method

Predicting future traffic is a critical task in intelligent transportation systems.Traffic predicting includes two parts:traffic collection and predicting.The traffic collection method is used to collect traffic in real time,and the traffic predict uses the collected historical traffic data to predict future traffic.Traffic predicting can provide decision support for traffic control and management,while ensuring the efficiency,safety and sustainability of traffic operations.We propose a new traffic collection and prediction method by analyzing some existing problems in the two processes of traffic collection and predictingAiming at the problem that the traffic collection method based on the Internet of Vehicles environment is difficult to balance the privacy protection performance,collection accuracy and storage communication overhead,we propose a traffic collection method that uses cuckoo hashing technology to reduce storage requirements.The size of the hash table can be adjusted dynamically,and the characteristics of low hash collision solve the above problems.We evaluate the proposed method from three aspects:privacy protection performance,acquisition accuracy,and storage communication overhead.It proves that our method can improve the accuracy of collection and reduce the storage and communication overhead of entities without reducing the performance of privacy protection.The focus of research on traffic predicting methods is mainly on improving the accuracy of single-period predicting,while less consideration is given to the problem of low predicting accuracy over multiple periods.In order to improve the accuracy of multi-step prediction and improve the ability to deal with road emergencies,applying Directed Graph Convolutional Networks(DGCN)and Recurrent Neural Network(RNN)to traffic prediction,we propose a deep neural network model DGCN-GRU based on DGCN and Gated Recurrent Unit(GRU).To extract the directionality of road traffic transfer,we define a second-order adjacency correlation and use DGCN to fully extract spatial features.At the same time,considering the time series of traffic,GRU is introduced to extract the time characteristics of traffic data.Experiments compare the new method with methods based on Graph Convolutional Networks(GCN),GRU,and Temporal Graph(TG).The results show that the model in this paper has improved the prediction accuracy,and the prediction accuracy in multi-period tasks is high and stable;the prediction trend is closer to the real traffic changes.In addition,the new method has a better fitting ability for sudden changes in traffic caused by road emergencies.We implement a simulation system based on parallel computing to verify the method proposed above,and considers the problem of high training time.The system includes functional modules such as traffic collection,traffic data processing,and traffic predicting.The system test results show that the system can adjust the parameters according to the needs of users to meet various needs of users.