Design And Implementation Of City Short-term Traffic Flow Forecast And Visual Analysis System

Timely accurate traffic flow prediction is the prerequisite and foundation for urban traffic control and guidance,and it is an indispensable part of the Intelligent Transportation System(ITS).Forecasting the traffic flow is crucial for governments,road users and private sectors.Due to the highly non-linearities and complex patterns of traffic data,most of the existing short-term traffic flow prediction methods lack the ability to model the temporal and spatial dependencies of traffic data,so it is difficult to yield satisfactory prediction results.In addition,model-based methods are often used in traditional transportation systems which lead to problems,such as redundant system construction,hard to develop and high prediction costs.Therefore,this paper studies and implements a city short-term traffic flow prediction and visual analysis system based on neural network.The main work is as follows:(1)In order to capture the spatial pattern,the traffic network is defined as a dynamic undirected graph and the graph convolutional network employing spectral approaches is used as the feature extraction network in this paper.A selective attention mechanism is proposed to dynamically adjust the impacting weights between input sequences,and reduce the influence between nodes with irrelevant locations.(2)This paper summarizes the temporal properties of traffic data into three categories,consisting of closeness,period and trend,and models these properties with three independent components respectively.Using the convolutional neural network to capture the temporal dependencies of traffic data is 10 times faster than traditional methods using recurrent neural network.(3)In this paper,we propose a spatial-temporal residual convolution block with complete convolutional structures.It captures the remote dependencies of a region or even a city by stacking multiple spatial-temporal residual convolution blocks and uses residual units to ensure the training effect of the deep network.Experiments on the Caltrans Performance Measurement System(PeMS)demonstrate that the proposed model outperforms the baselines.Based on the studies mentioned above,this paper designs and implements a traffic flow visual analysis system which mainly contains three functional modules:the traffic flow preprocessing module,the traffic flow prediction module and the traffic flow visualization display module.The microservice architecture is used to resolve the problem of heterogeneous calls in this system,ensuring the high availability and high scalability.With the above functions,this system can guide users to make correct travel decisions,improve the efficiency of traffic travel,and alleviate problems such as traffic congestion and the waste of resources.Finally,the stability and effectiveness of the system are verified through system testing.