Short-term Traffic Low Prediction Based On Deep Learning

With the rapid development of the transportation industry,traffic data is continuously increasing.Scholars hope to use these data to serve urban traffic,so,Intelligent Transportation System becomes the research direction of the future traffic system.As a research hotspot in the field of Intelligent Transportation,short-term traffic flow prediction is of great significance for traffic diversion and path planning.In order to study short-term traffic flow prediction,this paper constructs a novel prediction model based on deep learning under multi-mode spatiotemporal features.The model uses three components to model the close,daily,and weekly characteristics of traffic flow.And each component combines graph convolutional calculation and recurrent neural network to capture the spatiotemporal correlation characteristics of traffic data directly.Besides,considering the adverse impact of missing data in traffic data on shortterm traffic flow prediction,the model based on bidirectional gated recurrent neural network is constructed to study the filling strategy about traffic missing data,further analyzing short-term traffic flow prediction.The main research work and contributions of this paper are as follows:1.According to the spatial-temporal characteristics of traffic flow data,constructing the method based on deep learning is to study short-term traffic flow prediction.First,considering the characteristics of the actual road network,using the graph models the network topology to find the correlation between different locations.Next,considering the spatial-temporal correlation of traffic flow data,we combine the recurrent neural network with graph convolutional calculation to build graph convolutional recurrent neural network as the component approach,which can simultaneously capture spatiotemporal dependencies.Finally,on account of multi-mode characteristics in traffic flow,the model with multiple components is established to predict the short-term traffic flow and fully analyze the change of traffic flow.2.The lack of traffic flow data makes a negative impact on the analysis of short-term traffic flow and the deep mining of traffic data.In order to further deepen the short-term traffic flow prediction,this paper proposes a method based on recurrent neural network to fill the missing data in traffic flow sequences,which can guarantee the integrity of data as much as possible.The method simultaneously considers the historical and future associated information of the missing data by introducing a bidirectional gate recurrent neural network with the double-layer reverse structure,to achieve the effective recovery of traffic flow sequences.Finally,the paper uses the real traffic data set for experimental verification.The experimental results show that our prediction model can adequately consider spatiotemporal characteristics and achieve good prediction results.Meanwhile,the missing data filling method proposed in this paper has performed good filling effect under certain conditions,and the filled traffic flow data also has a better performance in the short-term traffic flow prediction.