Urban Grid Traffic Flow Prediction Based On Community-3DCN

In recent years,with the leapfrog development of scientific and technological strength,people’s quality of life has been significantly improved.With the increasing traffic demand and fixed and limited road space,traffic congestion has gradually become a worldwide problem.But at the same time,the quality and accuracy of traffic information collection are also improving.Real-time and accurate data can be obtained from the road traffic network,and road congestion and traffic flow prediction can be evaluated by combining the proper deep learning model.In this paper,based on the temporal and spatial correlation and heterogeneity of traffic flow data,urban regional information and traffic flow value are combined.Based on three-dimensional convolutional neural network,spectral clustering algorithm is used to divide urban areas and construct a Community-3DCNN model.The composite 3DCNN was applied to the regional traffic flow prediction,and the 2D convolution layer was superimposed to explore the spatial information,and then the recalibration block was embedded to capture the contribution of features to different grid areas and capture the spatial heterogeneity.Finally,on the basis of the model,the spectral clustering algorithm is applied to the classification prediction of urban grid region.The Community-3DCNN model proposed in this paper was applied to the given Taix BJ dataset and NYCBike dataset,and compared with the traditional model and the non-community model.The results showed that the composite 3DCNN model had the best performance on the two data sets,and the prediction errors RMSE values were 16.09 and 5.81,respectively,while the prediction errors RMSE values of the Community 3DCNN model proposed in this paper were 16.02 and 5.77 for the urban congestion area in the two data sets.Therefore,The prediction performance of congestion area alone is better than that of the whole.On the whole,based on the composite three-dimensional convolutional neural network model,this paper proposed the Community-3DCNN model to fully extract the spatial and temporal correlation and heterogeneity information from the traffic flow data,and combined with the spectral clustering algorithm to divide urban areas,so as to achieve accurate prediction of the future traffic flow in congested areas within a small error range.