Deep Representation Learning And Traffic Flow Prediction Research Of Dynamic Spatio-temporal Data

The first measure to build a smart city is to build an intelligent traffic system.Traffic flow prediction is the key to realize it.With the rapid development of data sources,more and more data can be obtained.How to make full and reasonable use of massive data to accurately predict traffic flow is an urgent task.However,traffic data has complex dynamic temporal and spatial dependence and is also affected by external factors such as weather and holidays,which brings great challenges to traffic flow prediction.Based on the traffic flow data of Luohu District in Shenzhen city,this paper firstly preprocesses the data,deals with the problems of missing values and outliers in the data,and screens the features by random forest method.Secondly,the traffic flow data are analyzed in time and space.Then the traditional time series model ARIMA model is established and its correlation analysis and visualization,and the future data is predicted.Using support vector regression model and the convolution of the neural network method for forecasting of traffic flow,and designed a improved depth based on spatial and temporal characteristics of emergent against network learning model,the converged network to integrate from different areas of the external factors include whether for the weekend,whether for holidays,rainfall,temperature and other factors and data input to the encoder.Finally,the mean square error of the improved generative adversarial network model based on spatio-temporal characteristics for traffic flow prediction of the next day is 1.33,which is far lower than 44.67 of ARIMA model,35.42 of support vector regression model and 12.56 of convolutional neural network.Therefore,the improved generative adversarial network deep learning model based on temporal and spatial characteristics has a significant effect on traffic flow prediction.In this paper,a traffic flow prediction model is established based on the traffic flow data of Shenzhen city.The experimental results show that the improved generative adversarial network model based on the spatio-temporal characteristics has better prediction effect.