Analysis And Prediction Of Short-Term Traffic Flow In Road Network Based On Deep Learning

Urban transportation is one of the important indicators to measure urban modernization management.Traffic congestion is becoming more and more serious and traffic accidents occur frequently,which affect the efficiency of urban operation and residents' travel experience.Therefore,urban traffic management has gradually become the focus of attention.In the information age,it is expected to improve the level of urban traffic management and the quality of public services through technical means such as data analysis and data mining based on big data.In order to solve or improve the related problems in the field of transportation,this paper focuses on short-term traffic flow prediction,vehicle driving trajectory prediction,traffic congestion analysis,etc.The main research results obtained include:(1)A hybrid LSTM neural network structure is proposed,and the hybrid LSTM neural network structure and parameters are deeply tuned for different traffic conditions in the actual road network,then compared with other models or algorithms.The experimental results show that the maximum absolute error between the actual value of road traffic and the predicted value is 0.65,and the maximum relative error between the actual value of road traffic and the predicted value of intersection traffic is-4.00%.The accuracy of prediction is ideal,compared with other models and LSTM before optimization,the accuracy of the hybrid model has been significantly improved,and the running time of the model accords with the real-time performance of short-term traffic flow prediction.(2)The Markov chain weighting model is used to predict the motion trajectory of the moving object.After matching the simulated road network with the actual road network,the road network is cut to generate a smaller dimension transition probability matrix,which effectively solves the problem that the traditional low-order Markov model has low prediction accuracy and the high sparse rate of the high-order Markov model matrix will lead to a sharp increase in computation.(3)Real-time traffic flow prediction is carried out by regression analysis,and the optimal regression model is obtained by comparing the accuracy of ordinary linear regression analysis,regression analysis based on random gradient descent and local weighted linear regression model.The results of the optimal regression analysis model are stepwise regressed by using different statistical analysis methods,and the optimal simplified model with fewer independent variables is obtained through multiple iterations,the accuracy of regression equation of original model and regression equation of simplified model is verified.Finally,the relative importance of the interaction between intersections is calculated for the subsequent research on the correlation analysis of traffic flow and the causes of congestion.(4)By matching the simulated road network with the actual road network,more road attributes can be obtained,such as the number of lanes,the length of roads,etc.According to the road attribute and the original data attribute,the corresponding road indicators are calculated.Combined the actual traffic demand with the original data attribute,the traffic density,road saturation,peak hour coefficient and other indicators are used to judge whether the road is congested or not and the level of road service.Finally,based on the original data set and experimental results,the data visualization is completed,and the traffic flow heat map and the intersection traffic flow heat map based on the simulated road network are generated,and different colors are used to mark the congestion degree of each section.