Short-Term Traffic Flow Prediction Based On Deep Learning

With the rapid development of urbanization,the number of vehicles is increasing sharply,and traffic congestion appears in cities more and more frequently.Intelligent transportation system can effectively manage and control the ground transportation system.Short-term traffic flow prediction is a core technology in intelligent transportation system(ITS).The purpose of short-term traffic flow prediction is to predict the number of vehicles within a given time interval on the basis of historical traffic information.Accurate realtime traffic flow prediction can provide road users with information to optimize their travel plans and reduce their associated costs.According to the results of traffic flow prediction,the authorities can adopt advanced traffic management strategies to alleviate traffic congestion.More and more researchers begin to pay attention to short-term traffic flow prediction problem.In the past decades,researchers have proposed many short-term traffic flow prediction methods.The traditional methods for traffic flow prediction are parametric approaches.Due to the stochasticity and nonlinearity of the traffic flow,parametric approaches cannot describe traffic flow precisely,which makes the stability of its prediction ability very limited.Many researchers try to use nonparametric methods to predict short-term traffic flow.Owing to the ability of dealing with high-dimensional data,flexible model structure,strong generalization and learning ability of deep learning methods,many deep learning models and structures were applied for traffic flow prediction.In this paper,we study the short-term traffic flow prediction problem based on deep learning.The main research contents and contributions of this paper can be summarized as follows:1)We review and summarize the previous studies on short-term traffic flow prediction at home and abroad.Various mainstream short-term traffic flow prediction models are systematically summarized and compared.By analyzing various models and strategies,we sort out the problems of current short-term traffic flow prediction methods.2)We introduce the related theories and technical foundations of short-term traffic flow prediction and deep learning.We introduce the definition of short-term traffic flow prediction,the collection methods of traffic flow data,the preprocessing of traffic flow data and the evaluation indexes used to evaluate the effect of traffic flow prediction model.At the same time,we have made a simple combing of the basic concepts of deep learning,the training process of deep learning,convolutional neural network and recurrent neural network.3)We propose a short-term traffic flow prediction framework based on deep bidirectional long short-term memory network model.In view of the problems existing in some short-term traffic flow prediction models,such as inability to mine the deep features of traffic flow,inability to effectively utilize time-sensitive traffic flow data and inability to consider the impact of other characteristics of traffic flow,we propose a deep bidirectional long short-term memory network model.By combining the long short-term memory network,residual connection,deep hierarchical structure and bidirectional traffic flow,it can extract the deep features of traffic flow and make effective use of time-sensitive traffic flow data.In addition,we take precipitation information into consideration when predicting traffic flow.The characteristics of traffic flow under different precipitation conditions can be understood by using both traffic data and precipitation data.4)We propose a framework for urban traffic flow prediction based on spatial-temporal features.Traffic flow prediction is a typical temporal and spatial process.Thus,it is not appropriate to consider the temporal features of traffic flow or the spatial features of traffic flow separately.Therefore,as for a challenging application scenario branch of short-term traffic flow prediction: urban traffic flow prediction,we propose a deep urban traffic flow prediction framework DST based on temporal and spatial features.In our framework,we use a local convolutional neural network model(local CNN)to extract spatial features,and a long shortterm memory network model(LSTM)to extract temporal features.By combining the temporal and spatial features of traffic flow,our model has achieved good performance.Overall,based on deep learning methods,this paper makes full use of the model ability of deep learning to study the short-term traffic flow prediction problem.Besides,extensive experiments prove that our shortterm traffic flow prediction frameworks can effectively predict traffic flow and have a certain degree of robustness and flexibility.