Foundation model estimates appliance energy use from home meters
Ask for Any Appliance: A Prompt-Programmable Foundation Model for Non-Intrusive Load Monitoring
Machine Learning
Summary
Measuring how much power each appliance in a home uses usually requires separate models or specialized sensors, which can be costly. The authors created a single flexible model that can estimate the electricity use of any appliance requested by name, using only the overall power usage measured at the home. Their model is trained on large diverse data and uses natural language prompts to identify appliances, needing no extra specialized models. This approach showed good accuracy across several real-world datasets and can improve with example activations without retraining.
What this means in practice
- •For smart home device engineers: Develop smart home energy monitors that identify appliance usage from aggregate meter data using flexible natural language prompts.$Commercial implications: Enables energy management products to support many appliances without building separate models, reducing development cost and expanding coverage.
- •For building energy management teams: Use a single model to monitor multiple appliance types from whole-building meters for energy efficiency and fault detection.
Authors
Xudong Wang, Jiacheng Cui, Junyu Xue, Tongxin Li, Guoming Tang
Abstract
Non-intrusive load monitoring (NILM) estimates appliance-level consumption from a whole-home meter, but appliance-specific models and fixed output inventories make coverage costly to extend. We present FM4NILM (Foundation Model for NILM), a single prompt-programmable model that estimates a requested appliance's power trajectory from aggregate measurements, a natural-language description, and optional activation exemplars. A lightweight cadence-aware transformer is pretrained by masked reconstruction on 645k sequences from seven public corpora spanning 1-60 s sampling intervals, then aligned with appliance requests using observation-masked losses for partially labeled households. A Bernoulli-lognormal decoder separates activity detection from conditional power estimation. On held-out households and time periods from REDD, UK-DALE, and REFIT, one frozen text-prompted model serves twelve appliance-corpus requests, achieving 0.556 event F1, 0.625 AUPRC, and the lowest active-window MAE (251.8 W) among seven appliance-specific baselines. Streaming score aggregation raises event F1 to 0.582 with a 60 s aggregation delay. In a separate category-held-out evaluation, adding ten activation exemplars raises microwave AUPRC from 0.132 to 0.214 without parameter updates. Input-intervention ablations probe the model's dependence on appliance requests and aggregate measurements. These results demonstrate competitive disaggregation with one shared model and support extending appliance coverage through prompts and examples rather than additional specialist networks.