Highway Congestion State Prediction Study Based On Multi-source Data Fusion

The increasing number of motor vehicles and the increasing demand for transportation have brought serious challenges to the road capacity and service level,resulting in frequent traffic congestion.Therefore,it is of great significance to study,analyze and accurately predict the operation situation of expressway.The traditional congestion prediction research mostly uses a single data source,with uneven data quality and lack of certain redundancy and complementarity.Through the multi-source data fusion of expressway,the traffic state can be better restored,and the fused traffic flow parameters can be predicted using the deep learning model.The early warning method of expressway congestion state based on multi-source data fusion is studied,and reasonable control measures are formulated for the expressway management department It is of great significance to provide accurate traffic information for the public and improve the operation efficiency of expressway network.Based on the multi-source detection data of expressway,the paper extracts the traffic flow parameters of expressway,proposes the hierarchical analysis data fusion algorithm based on minimum variance,builds the traffic volume and travel time prediction model based on deep learning,proposes the identification and early warning process of expressway congestion,and realizes the early warning of expressway congestion.The paper preprocesses the expressway toll data,ETC gantry data and "two passengers and one danger" track data,and extracts the traffic volume and travel time data of different periods of the expressway section according to their data characteristics.After introducing the basic concept and development history of data fusion and analyzing and comparing the commonly used data fusion methods,this paper proposes the AHP data fusion algorithm based on the minimum standard deviation to fuse the traffic volume and travel time data.On the basis of traffic volume fusion,considering the factors that link traffic flow will be affected by its upstream and downstream links,a spatiotemporal influence matrix is proposed to help neural network learn the spatiotemporal characteristics of data.For the traffic volume data with dynamic,similarity,periodicity and spatiotemporal correlation,after analyzing the principle and process of Resnet residual neural network and Bi-LSTM network algorithm,the future traffic volume of the road segment is predicted by constructing the Resnet-BiLSTM combined traffic volume model.Comparing the algorithm proposed in the paper with the classical deep learning algorithm,MAE and RMSE have decreased to a certain extent,The prediction accuracy is effectively improved.After analyzing the attention mechanism and the principle and process of Transformer model,on the basis of travel time fusion,considering that the link travel time of a certain period of expressway will be affected by its front and back periods,the paper uses the jump expansion attention structure to learn the data characteristics of a long time scale without dividing the sequence segments.By constructing the Conv LSTM-Transformer travel time prediction model based on the self-attention mechanism to predict the travel time of expressway sections,a shortterm prediction model suitable for the travel time of expressway sections is obtained and good prediction results are obtained.The paper puts forward the early warning process of expressway traffic congestion,analyzes the causes of expressway traffic congestion,introduces the relevant expressway congestion evaluation indicators,and analyzes the impact of road alignment and traffic structures on traffic congestion and the relationship between traffic flow parameters and traffic congestion.After analyzing the relationship between traffic flow parameters and congestion evaluation indicators,it is determined to use travel time index(TTI),traffic flow density and "two passengers and one danger" vehicle speed to comprehensively determine the traffic operation status and obtain good early warning results.