Short-term Traffic Flow Prediction Based On Complex Network

With the rapid development of cities,traffic is getting busier and busier,and various kinds of traffic congestion and traffic accidents appear,making rational and efficient traffic planning a challenge.As an innovative and cost-effective transportation application,Intelligent Trans-portation System(ITS)is constructed to solve urban traffic problems.Trustworthy repair of the data quality of traffic flow and quality of data are important foundations for building ITS.Traffic flow prediction is an important part of the ITS.Therefore,traffic flow prediction is es-sential and challenging in traffic management and research.Based on the existing research,the machine learning algorithm and deep learning algorithm were improved and optimized in this paper.The complex network approach was combined to analyze the data.The traffic flow data restoration,complex network construction work,and high accuracy traffic flow prediction under different road conditions with different data sets were studied.The main results are as follows:(1)In order to better repair the missing values appearing in the traffic flow data,an adap-tive hybrid order exponential smoothing algorithm with residual correction is proposed.The parameters of the proposed algorithm are optimized using a quantum particle swarm algorithm to improve the repair accuracy.The algorithm’s effectiveness is verified by example analysis through experiments with different missing ratios of different data sets.The method provides accurate access to traffic flow data.It lays the foundation for subsequent traffic flow complex network construction,analysis,and prediction.(2)A complex network construction method of traffic flow based on gray correlation is proposed for the clustering problem of different traffic flow observation nodes.The Louvain algorithm is used to detect the communities of the constructed complex network and obtain the clustering results.A posteriori analysis of the community detection results shows apparent differences between groups,effectively distinguishing traffic flow observation nodes with dif-ferent changes.This method guarantees accurately obtaining the clustering results of different traffic flow observers.It lays a foundation for the subsequent construction of equivalent training sets and traffic flow prediction.(3)In order to improve the accuracy of traffic flow prediction,a traffic flow prediction model with a multi-periodic feature Exattention-GRU algorithm is constructed.The model uses a Gated Recurrent Unit(GRU)as the main body of the network,and the GRU is improved by introducing an external attention mechanism and residual structure to enhance the network performance.The improved GRU network is stacked and fused according to different cycle features to allow the model to learn different traffic information and improve the network accu-racy.The model is trained using an equivalent training set to make full use of existing data and improve the prediction accuracy; the RAdam algorithm solves the network parameters to obtain better accuracy.In verifying the validity of the model with actual data,a comparison experi-ment is conducted for the training effect of the equivalent training set.The amount of data in the training set is expanded by using the complex network method to build the equivalent training set,and the noise in the training set is effectively reduced compared with the way of using all nodes.The multi-periodic feature Exattention-GRU algorithm has better generalizability than several algorithms for different datasets.It can achieve good prediction results by adapting to highly complex traffic conditions.