Web platforms rely heavily on multivariate time series such as service telemetry, web traffic logs, and sensor streams served via APIs. To save storage and bandwidth, these series are often archived at coarse granularity, which prevents downstream tasks like autoscaling, alerting, and billing from accessing the fine‑grained forecasts they need. We therefore introduce Multivariate Time Series Super‑Resolution Forecasting (MTS‑SRF), which aims to predict fine‑grained futures from coarse histories, and propose TimeMagnifier to tackle three core challenges.
Fourier Anchor fits each input window to a trend‑Fourier structure with sparse cross‑variable gating, mitigating data sparsity. De‑noise Only refines ambiguity along the resolution axis rather than the noise axis, extending the grid level by level. Banded DTW Alignment corrects cross‑resolution misalignment against the coarse forecast using dynamic time warping.
On the first MTS‑SRF benchmark, covering ten datasets from five domains, real archived cross‑resolution pairs, strengthened baseline adaptations, and five‑seed significance tests, TimeMagnifier outperforms 15 baselines on every dataset, resolution coefficient, and metric, cutting average MSE by 12.3%. Its De‑noise Only and DTW stages also serve as plug‑ins that further reduce the MSE of existing forecasters by 5.1%–9.0%. The code will be released publicly.
Review