Federated Non-Intrusive Load Monitoring for Privacy-Preserving Demand Response
Federated Non-Intrusive Load Monitoring for Privacy-Preserving Demand Response
Description
Details
Context and Problem Statement
Non-intrusive load monitoring can infer appliance-level consumption from aggregate household meters, but high-resolution load curves may reveal occupancy and daily routines. Centralized learning therefore creates significant privacy risk.
Research Question
What disaggregation-performance loss must be accepted to measurably reduce inference of household occupancy patterns under federated learning and smart-meter computing constraints?
Proposed Approach
Train sequential disaggregation models federatively across simulated households, apply differential privacy to model updates, and audit privacy using attacks designed to infer occupancy or appliance behavior. Relate retained model utility to achievable demand-response benefits.
Expected Contribution
An empirical performance-privacy curve for NILM and an assessment of demand-response value retained at different privacy levels.
Expected Prototype
A federated demonstrator with multiple embedded clients, appliance-level usage outputs, and privacy-budget reporting.
Datasets
UK-DALE, REFIT, and measurements from an instrumented domestic testbed.
Challenges
Appliance heterogeneity, low smart-meter sampling rates, utility degradation under privacy, and social acceptance of fine-grained monitoring.
Research Question
Innovation
Expected Deliverable
Technologies
Required Skills
- Signal Processing and Time Series
- Federated Learning and Privacy
- Embedded Systems
- Electricity Markets and Demand Response Fundamentals
Datasets
- UK-DALE
- REFIT Electrical Load Measurements
- Measurements from an instrumented domestic testbed