Multi appliance load monitoring improves accuracy with stable training method
Multi-Appliance Non-Intrusive Load Monitoring via Label-Preserving Aggregate Recomposition and Prediction Consistency
Machine Learning
Summary
Electricity monitors often struggle to tell which appliance is using power when looking at the total household usage. The authors show a new training method that keeps appliance power patterns but swaps out background noise, helping models learn better. This approach makes predictions more consistent and accurate across different homes without extra work when used. Their method reduces errors in estimating power use for multiple appliances in real homes.
What this means in practice
- •For energy data scientists: Train models for better appliance-level power usage estimation across different households using label-preserving recomposition to reduce errors.
- •For smart home device engineers: Improve accuracy of multi-appliance energy monitoring devices without increasing inference complexity by applying the proposed training methods.$Commercial implications: Enables more accurate and reliable energy disaggregation in smart home products, improving device market appeal.
Authors
Jiangfeng Liu, Yanfang Fan
Abstract
Non-intrusive load monitoring (NILM) estimates appliance power sequences from aggregate power, but models trained on source households commonly lose accuracy in unseen households. Aggregate power also contains loads from other appliances and measurement error, so predictions may depend on the residual background that co-occurs with source-household targets. Time-aligned submetered measurements and the additive decomposition of aggregate power expose a relation unused by window-wise supervision: an aggregate window can be recomposed by replacing only its residual background while preserving all modeled target-appliance power sequences pointwise. We combine label-preserving aggregate recomposition with prediction consistency. Both windows receive complete power and operating-state supervision. For each appliance, disagreement between the two power predictions is penalized only when both satisfy a fixed reliability criterion and only to the extent that it exceeds a fixed margin. The proposed method is implemented using a multi-appliance architecture with two-stage shared-to-specific mixture-of-experts routing. On REDD, UK-DALE, and REFIT, the proposed method lowers appliance-averaged mean absolute error relative to single-window training from 14.75 to 13.14 W, from 8.88 to 8.51 W, and from 15.83 to 14.55 W. Label-preserving aggregate recomposition and prediction consistency are used only during training, and add no inference-time module or parameter.