Sim-MSTNet: sim2real based Multi-task SpatioTemporal Network Traffic Forecasting
Hui Ma, Qingzhong Li, Jin Wang, Jie Wu, Shaoyu Dou, Li Feng, Xinjun Pei
TL;DR
Sim-MSTNet tackles data scarcity and task imbalance in cellular traffic forecasting by combining sim2real data generation with domain randomization and a bi-level optimization to reweight samples. A cutting-plane algorithm solves the constrained bilevel problem, enabling effective integration of synthetic and real data. The architecture blends spatiotemporal feature extraction with cross-task attention and a dynamic loss weighting strategy to enable soft parameter sharing and mitigate negative transfer. Empirical results on Milano and Trento datasets show consistent improvements over state-of-the-art baselines in both single-task and multi-task scenarios, highlighting improved generalization under data limitations.
Abstract
Network traffic forecasting plays a crucial role in intelligent network operations, but existing techniques often perform poorly when faced with limited data. Additionally, multi-task learning methods struggle with task imbalance and negative transfer, especially when modeling various service types. To overcome these challenges, we propose Sim-MSTNet, a multi-task spatiotemporal network traffic forecasting model based on the sim2real approach. Our method leverages a simulator to generate synthetic data, effectively addressing the issue of poor generalization caused by data scarcity. By employing a domain randomization technique, we reduce the distributional gap between synthetic and real data through bi-level optimization of both sample weighting and model training. Moreover, Sim-MSTNet incorporates attention-based mechanisms to selectively share knowledge between tasks and applies dynamic loss weighting to balance task objectives. Extensive experiments on two open-source datasets show that Sim-MSTNet consistently outperforms state-of-the-art baselines, achieving enhanced accuracy and generalization.