A Predictor Based on Parallel LSTM for Burst Network Traffic Flow.

The network traffic prediction is a key step for service quality control in computer network. Aimed at the problem that the performance of the traditional prediction method significantly degrades for the burst short-term flow, this paper proposed a double LSTM architecture, one of which acts as the main flow predictor, another as the detector for the moment the burst flow starts. The two LSTM unit can exchange their internal state's information, and the predictor uses the detector's information to improve the accuracy of the prediction. To train the offline double LSTM architecture, a Depth-Backstep algorithm is put forward. To use the architecture to perform the online prediction, a pulse series is used as a simulant of the burst event. A simulation experiment is designed to test performance of the predictor. The results of the experiment show that the prediction accuracy of the double LSTM architecture is significantly improved, compared with the traditional single LSTM architecture.