Multi-Step Traffic Prediction for Multi-Period Planning in Optical Networks

TL;DR

This work addresses proactive provisioning in elastic optical networks by predicting traffic over a multi-step horizon using an ED-LSTM encoder–decoder. It demonstrates a four-step-ahead horizon ($u=4$, i.e., $2$ hours) and shows how multi-step predictions can guide two RSA heuristics, MAD-SA and MMD-SA, to reduce service disruptions while managing spectrum efficiency, outperforming a single-step baseline SSD-SA. Experiments on the Abilene dataset show zero blocking across schemes and disruptions reduced by up to about $34\%$, with MMD-SA offering the strongest disruption reduction at the expense of higher overprovisioning. The results highlight the practical value of long-horizon ML-driven planning for adaptive, QoS-aware optical network provisioning.

Abstract

A multi-period planning framework is proposed that exploits multi-step ahead traffic predictions to address service overprovisioning and improve adaptability to traffic changes, while ensuring the necessary quality-of-service (QoS) levels. An encoder-decoder deep learning model is initially leveraged for multi-step ahead prediction by analyzing real-traffic traces. This information is then exploited by multi-period planning heuristics to efficiently utilize available network resources while minimizing undesired service disruptions (caused due to lightpath re-allocations), with these heuristics outperforming a single-step ahead prediction approach.

Figures (3)

• Figure 1: A generic ED-LSTM architecture.
• Figure 2: ED-LSTM training evolution vs. MSE loss for $D_1$, $D_2$, and $D_3$ Abilene nodes.
• Figure 3: MMD-SA and MAD-SA examples, given the spectrum predicted for each time interval $t+i$ and for each connection $c_i$, with $u=4$.