In order to solve the problem of traffic congestion on urban main roads,a deep learning based method for predicting the level of traffic congestion on urban main roads is proposed.A Katz similarity matrix is established for the traffic network of urban main roads.It can preserve the structural features of the road network and obtain traffic flow data of urban main roads.The local sensitivity discriminant analysis model is used to map the traffic flow data to the low dimensional manifold,obtain the optimal projection matrix,and extract the traffic characteristics of urban main roads.Combining the recurrent neural network model(RNN)and the long-term and short-term memory network model(LSTM),the recurrent neural network model for long short-term memory(RNN-LSTM)model is designed to solve the vanishing gradient problem.The network is inputted the traf-fic characteristics of urban main roads,and outputs the prediction results of traffic congestion level of urban main roads after training.The experimental results show that the prediction accuracy of proposed method is between 0.8～0.98,and the aver-age prediction time is 24.74 ms,which has certain application value.
关键词
深度学习/城市主干道路/长短时记忆循环神经网络模型/Katz相似度矩阵/交通拥堵水平预测
Key words
deep learning/urban main roads/recurrent neural network model for long short-term memory/Katz similarity ma-trix/prediction of traffic congestion level
维普
万方数据