Research On Short-time Traffic Flow Analysis And Prediction Based On Deep Learning

With the rapid development of urbanization,the rapid growth in the number of motor vehicles and the growing demand for public transport services been putting tremendous pressure to the existing transportation system.Serious traffic problems obviously reduce the operational efficiency of urban traffic and the comfort of commuters.In the field of traffic flow forecasting,although some classical forecasting methods have high computational efficiency,they neglect the highly nonlinear and dynamic nature of traffic data,and are only suitable for traffic flow forecasting with low precision and stable running state.Nowadays,the combination of deep learning and prediction model has achieved good results in mining deep and implicit spatio-temporal correlation in large traffic data.In this paper,combined with the spatiotemporal characteristics of traffic flow and the method of deep learning,a hybrid neural network model is constructed to predict traffic flow,so as to provide more accurate traffic flow prediction for traffic control,so as to slow down or avoid congestion in advance.improve the operation efficiency of urban traffic.The main research contents are as follows:(1)Based on the traffic flows of real road networks,the collected traffic flow data are analysed for spatio-temporal characteristics.The traffic flow data is processed to filter and fix outliers,to reduce the impact of data quality on the training and performance of the prediction model,and to investigate the spatio-temporal characteristics and correlation of the threeparameter distribution of the traffic flow,in order to lay a good foundation for the subsequent construction of the traffic flow prediction model.(2)The combined traffic flow prediction model(CN-BL)based on spatio-temporal characteristics is proposed based on the dependence and non-linear correlation of short-time traffic flow in spatio-temporal series data.The prediction model takes historical traffic flow data as input and uses a convolutional neural network to fully capture the high-dimensional spatial characteristics of the dynamics of traffic flow data at observation points;the obtained temporal feature vector is then injected into the two-way long-and short-term memory network to deeply learn the time-series features in the data,and the real road network dataset is used to compare the error with the model prediction values,demonstrating that the CN-BL model is more accurate than traditional machine learning models in terms of prediction accuracy is superior to that of traditional machine learning models.(3)To further improve the accuracy and convergence speed of short-term traffic flow prediction,a prediction model(CN-AM-BG)based on a combination of a convolutional neural network(CNN)and a bi-directional gated recurrent network(BiGRU)with an attention mechanism is proposed to predict short-term traffic flow.An attention mechanism is used to reduce the loss of features from historical data and increase the focus on important node data to improve the model’s ability to capture deeper feature relationships in the data.In addition,the impact of different feature combinations on the prediction accuracy of the model under univariate and multivariate conditions is explored.Multiple comparison experiments demonstrate that multivariate feature inputs are more effective in improving the prediction accuracy of the model than univariate inputs.As an important foundation for intelligent transport systems,it combines deep learning methods to analyse and predict urban road traffic flows,providing real-time and accurate traffic flow trends.This is important for developing traffic guidance and traffic control schemes and improving road safety.