Research On Traffic Flow Prediction And Visualization Method Based On Deep Learning

In order to ease the current situation of urban traffic congestion and make full use of the urban road infrastructure,the intelligent traffic system is widely used in urban traffic management,which provides guarantee for road traffic safety,and improves the traffic efficiency and intelligent level of urban traffic.With the continuous development of urban dynamic traffic information collection technology,it is possible to obtain real-time road traffic data in the urban road network in a timely manner.A large amount of traffic data provides data protection for the analysis of urban road traffic condition and the prediction model study.Traffic flow prediction plays a key role in modern traffic management and control,which is the premise of urban traffic induction.Accurate and real-time traffic flow prediction can better analyze urban road traffic conditions,which plays an important role in urban traffic network planning and road traffic optimization control.The advantage of LSTM neural network in deep learning in the processing of time series data makes it very suitable for predicting short-term traffic flow.In view of this,based on the instantaneous monitoring data processing at vehicle monitoring points,this paper constructs a LSTM network model to predict the traffic volume through a certain road section.In the experiment,the key parameters in the prediction model were adjusted and compared,and the influence of different parameter settings on the accuracy of model prediction was analyzed.This paper makes full use of the good learning ability of the deep learning model and designs a deep network hybrid model based on deep learning for short-term traffic flow prediction.The hybrid model is constructed by a layer of denoising autoencoder and two-layer restricted boltzmann machine,and the SVR method is used as the predictor.In order to verify the performance of the forecasting model,relevant experiments were conducted on the monitoring point data on weekdays and weekends respectively,and compared with other forecasting models.It is concluded that the prediction model of this paper is good and has better prediction accuracy than other forecasting models.This paper designs a visual display system based on traffic data.For instantaneous monitoring data,the traffic conditions of the monitoring points and road sections are described through a visualized view of hourly traffic flow during the week and day.For summary monitoring data,this article uses the layout of multi-attribute axes to visualize the relationship between the basic parameters of traffic flow.Therefore,it is more intuitive and profound to understand the traffic data,which can help to understand the nature of its data and the mining of potential laws.