HXAI introduces a hierarchical framework that delivers explainable load information to grid operators while preserving user privacy. The architecture consists of a local model and a zonal model. The local model runs inside the household’s secure environment, generating fine‑grained explanations that never leave the premises. The zonal model aggregates statistical summaries of these explanations and enforces differential privacy through a tunable privacy budget, limiting cumulative privacy loss from repeated operator queries.
Experiments on both simulated and real energy datasets show that HXAI provides useful insights for regional load management without exposing appliance‑level consumption. The results also reveal that preserving the semantic structure of explanations is more important than merely minimizing numerical error, because differential privacy perturbs values while semantic cues can remain intact. This framework offers a practical path to reconcile privacy and explainability in energy management.
Review: Deploying HXAI requires careful handling of local computational resources and dynamic privacy‑budget allocation, yet its hierarchical design opens new avenues for scalable, explainable operation of large‑scale energy systems.