HXAI Protects Privacy While Explaining Distributed Energy Demand

Summary

Managing electricity demand is becoming harder as renewable generation remains intermittent and household consumption is stochastic. Grid operators need detailed, decision-relevant explanations of consumption to manage peak loads and design responsive tariffs, but household-level transparency can expose sensitive information. The paper introduces HXAI, a hierarchical explainable AI framework with two layers: a local model that produces fine-grained explanations inside a secure private environment, and a zonal model that aggregates those explanations for grid-level analysis. Flexible privacy-budget management limits cumulative exposure across repeated operator queries. Experiments on simulated and real-world energy datasets show that HXAI can provide useful information for zonal load management while keeping appliance-level consumption local and never transmitting it to grid operators. The authors report that preserving the semantic structure of explanations, rather than only minimizing numerical error, is central to maintaining useful explainability under differential privacy.