Spectral temporal graph model improves sensor network forecasting accuracy
ST-GFM: Spectral-Temporal Graph Forecasting with Measured Perturbation Bounds and Masked Spectral Reconstruction
Machine Learning
Summary
Predicting sensor readings over time is tricky because how sensors relate and the signals they pick up can change. The authors present ST-GFM, a model that uses mathematical filters and attention techniques to track these changes more accurately. They also add a special training method to make predictions match the input frequencies better. Their model is proven to be stable under certain changes and performs better than other methods on real traffic sensor data.
What this means in practice
- •For traffic management teams: Improve traffic flow predictions by deploying ST-GFM to handle changing sensor connections and signal patterns more accurately.
- •For industrial iot operators: Use ST-GFM to enhance forecasting of sensor data in industrial networks where sensor relationships and signals vary over time.
Authors
Aashish Bohra, Vivek Vijay
Abstract
Graph forecasters for sensor networks are trained in one regime and deployed in others, where sensor coupling and signal frequency content both change. We split this spectral shift into an operator channel and a frequency channel, each with a dimensionless measure. ST-GFM (Spectral-Temporal Graph Forecasting Model) filters every input step with Chebyshev and bidirectional diffusion filters on one learned, non-symmetric operator and alternates them with per-node temporal attention. For the operator channel we prove Frobenius-norm perturbation bounds for both graph branches, show that the Chebyshev bound is attained to first order, and compute its constant from the trained weights and the measured operator norm: 17.9 on METR-LA and 21.5 on PEMS-BAY. Under random edge dropout the trained layer moves by 2.0% to 3.1% of this bound. Masked Spectral Reconstruction (MSR), a self-supervised loss active in the first 40 epochs, matches the log power spectrum of the encoder output to that of the input. For small log-spectral residuals on unclamped bins it is proportional to an Itakura-Saito divergence to leading order. Removing it raises the 60-minute MAE by 0.30 on METR-LA, close to the 0.33 lost when all three self-supervised losses go. Under one shared evaluation pipeline, the 60-minute MAE of ST-GFM is 3.5% below a retrained STAEformer on METR-LA and 13.0% below it on PEMS-BAY. Its 60-minute MAE is 3.0980.017 on METR-LA (five seeds) and 1.5760.006 on PEMS-BAY (three seeds), the lowest PEMS-BAY MAE among twelve published baselines.