Research And Application Of Short-term Traffic Flow Forecasting Method On Urban Roads

With the acceleration of China’s urbanization process,the number of urban vehicles has continued to increase,and urban traffic demand has shown a growth trend,which has caused a series of problems related to people’s livelihood,such as traffic congestion and traffic accidents,and has seriously hindered the high-quality development of the city.In the current urban intelligent transportation system,rapid and exact traffic flow forecasting is an essential prerequisite for traffic control in urban areas and it plays an essential role in the intelligent transportation system.Given the above background,this thesis analyzes the data characteristics of traffic flow,proposes two short-term traffic flow prediction methods under different traffic flow original data,designs and implements a short-term traffic flow prediction system.The main work is as follows:(1)A combined forecasting model of short-term traffic flow based on improved Bayes is proposed.Because of the poor prediction accuracy of a single model and the small range of application scenarios,the autoregressive integrated moving average prediction model,the gray prediction model,and the long short-term memory network prediction model are combined,and the Bayesian algorithm is introduced into the combined weight evaluation.The Bayesian calculation method of recent iteration weights uses the latest prediction errors of the sub-models to dynamically allocate combination weights,which improves the sensitivity of the combined model while speeding up the iteration efficiency of the algorithm.Then the PEMS traffic data set is used to test the proposed model.The experimental results show that the combined prediction model is superior to the comparison model in terms of indexes such as root mean square error,and has higher prediction accuracy.(2)A short-term traffic flow prediction model based on road time and space characteristics is proposed.Given the insufficiency of traditional forecasting models that it is difficult to capture the time and space relationship of traffic flow,the original traffic flow data of the target road is decomposed into trend data,residual data and fixed period data,and the time dimension of the trend data part is analyzed by the Gated Recurrent Unit.Feature mining and prediction,while using a convolutional neural network to analyze the relationship between the surrounding road traffic flow and the target road residual data,and perform feature mining and prediction on the spatial dimension of the residual data part.Finally,the trend data,residual data,and period data are summarized to complete the final forecast.This section uses the Pe MS traffic data sets on the proposed model was tested,and the test results show that the proposed traffic flow prediction model can effectively mine the spatial and temporal dimensions of the target road.Compared with the comparison model,it has higher prediction accuracy.(3)Designed and implemented a set of urban road short-term traffic flow forecasting systems.The system can provide the function of simultaneous online model training and prediction by multiple users.The overall architecture and system sub-modules of the traffic flow prediction system are designed,and the engineering implementation details of the main components of the traffic flow prediction platform are described.