Traffic prediction for PON based on pre-matched cross-port transfer learning

Based on the long short-term memory (LSTM) network, a temporal model of matching cross-port Transfer Learning (MCPT-LSTM) for predicting the traffic of PON is proposed and numerically studied, in which transfer learning is used for transferring a pre-established model of a source PON to a target one, effectively mitigating the challenges associated with inadequate data that typically hampers the accuracy of network traffic predictions within PON systems. Experimental evaluations employing operational PON network data underscore that the proposed model enhances the precision of traffic forecasting by a margin exceeding 20%.