Graph learning improves forecasting across different hierarchical views
Learning to Forecast on Multiple Spatiotemporal Hierarchies
Machine Learning
Summary
Forecasting data that changes over time and space often uses hierarchical groupings, like different regions and their subregions. The paper’s authors found that training forecasting models on several ways of grouping data, instead of just one fixed hierarchy, helps the model learn better patterns that work more generally. They also created an efficient method to ensure the predictions are consistent across these hierarchies without needing heavy computation. Their approach improves accuracy compared to models trained on a single grouping and matches leading forecasting methods.
What this means in practice
- •For urban planners: Improve forecasting of city-level and neighborhood-level patterns by jointly using multiple spatial and temporal groupings for better resource planning.
- •For energy grid operators: Enhance electricity demand forecasting accuracy by integrating various hierarchical spatial zones and time scales in predictive models.
Authors
Valentina Moretti, Cesare Alippi, Andrea Cini
Abstract
Hierarchical forecasting exploits aggregation structures that expose multiple views of the same underlying spatiotemporal forecasting problem, while ensuring coherence among forecasts. Individual series provide fine-grained information, while progressively coarser aggregates reveal temporal and spatial patterns at different scales. In the associated deep learning architectures, the hierarchy determines which aggregate series are available and how they are processed together with individual series (e.g., through reconciliation or message-passing operators). Different hierarchies thus provide distinct learning signals and inductive biases for forecasting the same collection of time series. Nonetheless, existing architectures are typically trained on a single, often fixed, structure. In this work, we propose a graph deep learning framework that leverages variations across hierarchies as multi-task regularization. We show that jointly learning across these structures can reduce over-specialization to any single hierarchy and improve forecasting performance. Indeed, training across hierarchies provides an inductive bias toward representations that are useful across different structures, and as such, generalize better. Making forecasts coherent with the aggregation constraints for each sampled hierarchy is computationally expensive: to complement the framework, we propose a novel operator-based on a top-down approach-that produces coherent forecasts with linear time and space complexity w.r.t. the hierarchy size. Experiments show that our approach improves forecasting accuracy over training on a single structure and is competitive with current state-of-the-art methods.