Road Network Congestion Situation Prediction Based On Deep Combination Modeling

With the rapid development of Chinese national economy,traffic congestion in major cities has become more and more serious,which has greatly affected people’s travel efficiency.In response to such problems,researchers use a variety of algorithms to predict the change trend of the road network speed,hoping to divert traffic based on the prediction results.However,the current traffic congestion prediction research is based on a single road segment,and for roads composed of multiple road segments the traffic prediction accuracy of the network is low,and the real speed situation of the road network cannot be well represented.Based on such problems,this paper abstracts the speed changes of the road network at different times into pictures,plays them frame by frame,and predicts the next frames through the first few frames.Based on this,this paper uses deep learning to mine the temporal and spatial characteristics of each section of the road network.A more accurate prediction of the road network speed status is mainly divided into the following three steps:1.Grid division of the road network: divide the road network with grids of a specific size and establish a road network state matrix.In this way,the characteristics of the road network can be preserved,and the relative position of the road sections can be kept unchanged.It can also effectively avoid the change of road section shape.2.Extract spatial features: take the gridded road network matrix as input,and use convolutional neural network to mine the spatial features of each road section.3.Extract the time series features: use the output data processed by the convolutional neural network as the input data of the recurrent neural network,mine the time series features of the road network through the LSTM neural network,and predict the final speed change trend of the road network.After building this model,the model is verified with the urban road data in the third ring road in Beijing as an example,and the prediction results of traditional neural networks,convolutional neural networks,and recurrent neural networks are compared.The experimental results show that the prediction results of this model are more accurate in both the dense road network and the decentralized road network,both of which are above 90%.Through multi-angle comparison with existing prediction models,the applicability and superiority of this model in predicting road network congestion are verified.