Research And Implementation Of Traffic Flow Prediction On Highway Network

Highway network is an important infrastructure that promotes national economic and social development.The construction of the highway network started late,but it has developed rapidly in China.At present,the highway mileage has exceeded 130,000 kilometers,and the total mileage ranks first in the world.Ensuring the safety and efficiency of the highway network plays an important role in the stable development of China's national economy and society.Traffic flow prediction is one of the important research content in the intelligent transportation.How to improve the accuracy of traffic flow prediction by using new technologies is a problem that increasingly concerned by scholars.This thesis mainly analyzes the data characteristics of traffic flow and studies the key issues of traffic flow includes get and preprocess the data,design and optimize the network model,and finally present the prediction of highway network traffic flow by using the deep learning neural network.1.As the original data of the traffic flow has a few features,the machine learning method has being used to expanse the multi-feature of the highway network data sets.In the multi-feature expansion stage,one proved time-series clustering method based on the hierarchy has been designed for extracting the space feature of the highway network.This method has improved the clustering effects of the traffic flow data and effectively distinguished the node set of the highway network with different flow change features.2.For the space-time characteristic of nodes,one deep network model that can predict the traffic flow combining convolution and LSTM has been designed.This model can effectively adding the local spatial characteristics of nodes to the real-time prediction of the network by using the characters of convolution and advantages of LSTM in timing processing.Experiments show that this model not only can guarantee the accuracy of the prediction on nodes that have a single spatial structure,but also the accuracy of the prediction on regional nodes that have a complex spatial structure is improved obviously.3.Improve the prediction model by using multiple features.This paper explores a general method that incorporating weather,date,geographic location and other factors into the prediction model,and improves the original prediction model by designing the auxiliary features of traffic flow.Experiments show that the prediction effect has been improved and the stability of the overall prediction effect of multiple nodes that in the highway network is higher than before under this improved model.This thesis has realized the highway network traffic flow prediction system through the multi-analysis of the traffic flow prediction tasks by using the prediction model and the highway network data of Sichuan province.The experimental results show that the proposed method is higher in the accuracy of prediction than the mainstream model,and the stability of the model is better.