Real-time traffic prediction for PON based on adaptive online learning

In this paper, we proposes an online learning-based Long Short-Term Memory (LSTM) model which dynamically updates model parameters in real-time at a lower computational cost. This makes it possible to quickly adjust to the data's dynamic changes. The usefulness and potential of the model in real-time prediction scenarios are demonstrated by the experimental findings, which show that it performs better in real-time training situations than typical offline training models.