| As emerging technologies such as the Internet of Things,cloud computing and artificial intelligence have been widely applied in the field of smart cities,people’s living standards and quality of life have been significantly improved.However,traffic congestion in the field of intelligent transportation is still a livelihood problem to be solved.Therefore,how to effectively solve road congestion,accurately predict traffic flow,provide travel guidance for the public and improve travel experience has become an important research topic for scholars at home and abroad.In this thesis,the most fundamental spatio-temporal features of traffic data are captured,the stacking model of single time mode is combined with graph convolutional neural network and cyclic time network,and the final prediction results are obtained by fully connecting and combining the three kinds of time dimension suites.In order to prevent the incomplete data set from affecting the accuracy of the prediction model,a bidirectional gated neural network was used to restore and fill the incomplete data.The main research methods and innovative contributions of this paper are as follows:1.Taking the features of traffic flow data as the entry point and extracting the data features of time series and spatial road network topology,a stacking model based on the fusion of graph convolutional neural network and cyclic neural network is proposed.The signal processing in the graph convolution calculation is used as the representation of the structural features of the road network,and the time slice data is introduced into the recurrent neural network for the feature memory of the historical time.Finally,the two models are structurally-stacked to obtain a single time suite prediction model.2.Considering the phenomenon that the traffic scene will change due to the periodic changes of the mass travel rule,three kinds of time suites are divided into the single mode time suite fusion prediction model built above,and the output results of the three sets of time suites are all connected with the weighted output of the layer,and the traffic flow prediction model of the multi-mode and multi-suite graph convolution cycle network is finally formed.3.In the process of the experiment,it was found that due to the incomplete data,the accurate prediction effect of the model was seriously affected and the model could not carry out in-depth feature capture.Therefore,this thesis proposes the incomplete data filling model based on BGRN,which combines the historical forward and future reverse to recover the incomplete data through the bidirectional gating unit.Finally,according to the real data set published by Didi as the basis for the experimental evaluation of the model,the data were processed by means of desensitization,normalization and dirty data cleaning,and then fed into the short-term traffic flow prediction model integrated by the graph convolution cycle neural network of the multi-mode and multi-time set proposed in this paper.After the experimental evaluation and comparison,It is proved that the proposed MSGCN does have better feature extraction performance.At the same time,the incomplete data based on BGRN is used to restore and fill the incomplete data,which proves that the effective recovery of the incomplete data can effectively help the prediction model to improve the model fitting. |
