Prediction of traffic fluctuation in telephone networks with neural networks

Neural networks with back-propagation learning algorithms are applied to the prediction of traffic fluctuation in circuit-switched telecommunication networks. The selection of the neural network structure and input/output musters are discussed in detail. The prediction of traffic with neural networks is simulated and then compared with other prediction methods. The results show that neural networks can reduce the prediction error extensively. This forecasting method is then combined with adaptive routing, and the simulation shows that the performance of telecommunications networks, the sum of end-to-end blocking probabilities for all node-pairs, can be improved.