Flow Prediction in Spatio-Temporal Networks Based on Deviation Correction Hybrid Model

Flow prediction of spatio-temporal network is a very challenging problem in intelligent traffic system. The nodes in the spatio-temporal network have spatial correlation according to the geographical topology, and the node flow has temporal correlation in different time intervals. The traditional neural network model is restricted by the network structure and numerical calculation method in the field of spatial-temporal network traffic prediction, the prediction results inevitably have errors, and the time cost of model training can be very large when the data scale increases. To address these issues, we propose a mixed prediction model based on deviation correction, which can optimize the prediction results by adjusting the weights of different prediction models. In the single model part, the traffic prediction model based on LSTM and LSSVM, We solve the problems of high time correlation and model training cost of nodes by integrating the periodicity characteristics of the proposed node traffic into the prediction model. When the time-varying graph is converted into a tensor, the spatial correlation between nodes is maintained. We use taxicab data in Beijing to evaluate our method. Experimental results show that our method is superior to the baseline method in two evaluation metrics.