EnergyMamba: An Uncertainty-Aware Graph-Enhanced Selective State Space Model for Energy Consumption Prediction

Researchers introduce EnergyMamba, a machine learning framework that combines graph neural networks with state-space models to predict energy consumption while quantifying prediction uncertainty. The system achieves 5% accuracy improvement over existing methods by simultaneously modeling spatial grid relationships and temporal patterns, with enhanced reliability during abnormal conditions like extreme weather.

EnergyMamba represents a meaningful advancement in energy forecasting by addressing two persistent limitations in the field. Traditional approaches treat energy prediction as isolated time-series problems, ignoring how consumption patterns propagate across interconnected grid regions. This framework explicitly models grid topology alongside temporal dynamics, creating a more holistic representation of real-world energy systems. The integration of uncertainty quantification through Adaptive Sequential Conformalized Quantile Regression adds practical value—energy operators need reliable confidence intervals around predictions, especially during extreme weather when forecasts become most critical and least accurate.

The research emerges from the energy sector's growing complexity. As grids integrate renewable sources and face climate-driven demand volatility, traditional statistical models prove insufficient. Machine learning adoption accelerated post-2020, but most implementations focus on accuracy metrics while ignoring prediction reliability. This work fills that gap by providing calibrated uncertainty estimates that adjust dynamically to distribution shifts—a crucial feature for real-world deployment where grid conditions change unpredictably.

For energy utilities and grid operators, the 5% accuracy improvement translates directly to operational efficiency gains and reduced reserve capacity requirements. Better predictions enable demand-side management, cost optimization, and renewable integration. The uncertainty quantification enables conservative decision-making during critical periods, reducing blackout risks. Developers benefit from the open methodology combining graph-enhanced models with conformalized quantile regression—techniques applicable beyond energy to water systems, transportation, and infrastructure management.

• →EnergyMamba integrates spatial grid topology with temporal dynamics for coupled spatiotemporal energy prediction modeling.
• →Adaptive Sequential Conformalized Quantile Regression provides dynamically calibrated uncertainty estimates under distribution shifts.
• →Framework achieves 5% accuracy improvement and 6% uncertainty quantification improvement over 15 baseline methods.
• →Validated on large-scale datasets from Florida, New York, and California demonstrating real-world applicability.
• →Reliable uncertainty quantification enables grid operators to optimize reserves and manage extreme weather scenarios.