Urban Taxi Traffic Flow Prediction Model Based On Deep Learning

There are a large number of taxis in the city,and the problem of taxi flow prediction is closely related to people's life,especially in first-tier and second-tier cities.The road congestion caused by too many taxis seriously affects people's work and life.This problem is especially obvious in rush hour for go and off work.If we can accurately predict the taxi flow in a certain area of the city,it will be of great guiding significance for the unblocking of traffic and the deployment of taxis.Now the deep learning models have powerful representation capabilities,deep learning algorithms can be applied in cross-cutting areas,such as urban computing.This topic benefited from the published taxi traffic data,in order to predict the future taxi traffic at a certain time.In the actual prediction scenario,the model also needs to consider external factors,such as temperature,rainfall,holiday,date,etc.which will affect the accuracy of the prediction.The relevant work and research results are as follows:1)Design the urban taxi flow prediction model: Two taxi flow prediction models are designed,which are based on convolutional neural network CNN module and recurrent neural network RNN module,graph convolutional network GCN module and recurrent neural network RNN module,the RNN module uses the BiLSTM model,and all have the spatiotemporal attention mechanism,in the meanwhile,taking into account external factors.2)Problem definition and data set construction: Respectively defined the taxi flow prediction problems under the two models.From the United States New York taxi website,we collected yellow,green,and Uber taxi travel data from April 2014 to September 2014,and collected external factors datas such as temperature,rainfall,holiday,and date during the corresponding period.Handled the corresponding data sets for the two models.3)Horizontal comparison experiment: In the order of traditional statistical time series prediction algorithm,machine learning regression algorithm,and deep learning algorithm,the two models in this topic are compared with the models used in the past traffic flow prediction,and analyzed the results.4)Longitudinal comparison and saliency experiment: Conduct a longitudinal comparison ablation experiment to remove the spatiotemporal attention mechanism ofthe two models in this topic to verify whether adding the spatiotemporal attention mechanism can improve the accuracy of model prediction.And further conduct significant experiments to verify the accidental of the experiment results.5)External influence factors and saliency experiment: Conduct ablation experiments.In the model based on the convolutional neural network CNN module and the recurrent neural network RNN module,the external influencing factors are not considered,incording to verify whether adding external influencing factors can improve the model prediction accuracy.Further conduct significance experiment to verify the accidental of the experiment results.6)Time series fine-grained experiment: Conduct timeseries fine-grained experiment,analyze how long the time series should be,so that the models is not too complicated and have a certain degree prediction accuracy.