Search papers, labs, and topics across Lattice.
This paper introduces STOIC, a novel framework that combines spatial-temporal graph neural networks (STGNNs) with tabular foundation models for accurate energy demand forecasting and uncertainty quantification. By reformulating spatial-temporal residuals into a tabular format, STOIC leverages in-context learning to calibrate prediction intervals without the need for task-specific retraining, effectively capturing the complex dependencies inherent in energy systems. Evaluated across five diverse benchmarks, STOIC demonstrates superior performance in providing reliable uncertainty estimates compared to existing conformal prediction methods.
STOIC outperforms traditional conformal prediction methods, delivering robust uncertainty estimates that are crucial for risk-aware decision-making in energy systems.
Accurate energy demand forecasting is essential for the reliable operation and planning of modern sustainable energy systems. Spatial-temporal graph neural networks (STGNNs) have recently achieved strong performance in point forecasting by jointly modeling temporal dynamics and relational dependencies across interconnected energy nodes. However, in real-world energy systems, accurate point forecasts alone are insufficient, as operators also require reliable uncertainty estimates to support risk-aware decision-making, grid stability, and operational planning under uncertainty. Conformal prediction provides a principled and model-agnostic framework for uncertainty quantification with statistical coverage guarantees, making it particularly attractive for safety-critical energy applications. However, existing conformal prediction approaches often fail to fully capture the complex spatial-temporal structure of energy systems. To address these limitations, we propose STOIC (Spatial-Temporal Graph Conformal Prediction with In-Context Learning), a novel framework that integrates graph-based forecasting with the zero-shot calibration capabilities of tabular foundation models. STOIC first generates point forecasts using an STGNN and subsequently reformulates spatial-temporal residuals into a tabular representation suitable for in-context learning. Leveraging a tabular foundation model, STOIC calibrates prediction intervals without task-specific retraining, effectively capturing both sequential and relational dependencies. We evaluate STOIC on five diverse benchmarks, including synthetic simulations as well as real-world electricity and district heating networks. Across all datasets, STOIC consistently outperforms existing conformal prediction baselines, delivering more reliable and robust uncertainty estimates for complex graph-structured energy time series.
