Complement And Forecast Of Traffic Flow Based On Fusion Model

Traffic flow information is an important basis for Intelligent Transportation Systems（ITS）and urban computing.Traffic flow prediction is an important task of intelligent traffic management.However,due to the loss of traffic flow data,traffic flow prediction models are often inefficient and difficult to carry out.Traffic flow data is a kind of time series data that contains spatial geographic location information.In the process of data collection,transmission,and storage,it is inevitable that data will be missing.How to effectively mine the Spatial-temporal features of traffic flow data and the correlation between the data,as well as to complete the missing traffic flow data,has become the key and important means to improve the accuracy of traffic flow prediction.Some traditional statistical methods can not achieve better prediction results,and the application of deep learning promotes the development of traffic flow data complement and prediction to a higher accuracy.For the task of traffic flow data complement and prediction,this paper improves and integrates some deep learning models,and integrates these two aspects of work.On the one hand,in view of the problem of the missing of traffic flow,we first make different assumptions about the missing of traffic flow.Then,the characteristics of traffic flow change with time and the correlation of traffic flow in different spatial areas are deeply analyzed.At the same time,a fusion data complement model based on multi-view attention mechanism is proposed by using different data correlation measurement methods,so that the problem of missing traffic flow data is effectively solved.On the other hand,based on the traffic flow data after the model complement,a 3D dilated convolution residual network based on the multi-view attention mechanism is used to forecast the traffic flow.The model will be integrated with the historical data which associated with the time to be predicted,to fully explore the Spatial-temporal features of traffic flow.At the same time,it deeply analyzes the channel characteristics of traffic flow,and improves the loss function of the model,which further improves the prediction effect and accuracy of the model.The above two aspects of work integrate relevant research on traffic flow complement and prediction,improve the accuracy of traffic flow forecasting with existing missing data,and expand the application of deep spatial and temporal prediction model in traffic flow data prediction.In order to verify the effectiveness of the data completion and prediction model,this paper uses the Beijing taxi trajectory dataset for experiments.In the experiment,the influence of model fusion and loss function on the data complement and prediction accuracy is analyzed in detail,and some results are visualized and analyzed.The experimental results show that our data complement method and prediction model can further improve the accuracy of traffic flow complement and prediction.