Benchmarking Active Learning for NILM

TL;DR

This work is the first to benchmark the use of active learning for strategically selecting appliance-level data to optimize NILM performance, and develops uncertainty-aware neural networks for NILM and then installs sensors in homes where disaggregation uncertainty is highest.

Abstract

Non-intrusive load monitoring (NILM) focuses on disaggregating total household power consumption into appliance-specific usage. Many advanced NILM methods are based on neural networks that typically require substantial amounts of labeled appliance data, which can be challenging and costly to collect in real-world settings. We hypothesize that appliance data from all households does not uniformly contribute to NILM model improvements. Thus, we propose an active learning approach to selectively install appliance monitors in a limited number of houses. This work is the first to benchmark the use of active learning for strategically selecting appliance-level data to optimize NILM performance. We first develop uncertainty-aware neural networks for NILM and then install sensors in homes where disaggregation uncertainty is highest. Benchmarking our method on the publicly available Pecan Street Dataport dataset, we demonstrate that our approach significantly outperforms a standard random baseline and achieves performance comparable to models trained on the entire dataset. Using this approach, we achieve comparable NILM accuracy with approximately 30% of the data, and for a fixed number of sensors, we observe up to a 2x reduction in disaggregation errors compared to random sampling.

Figures (17)

• Figure 1: Active learning (AL) loop showing the flow of data across the whole AL pipeline
• Figure 2: The mains power data is available for all the houses in the experiment (light blue color). The appliance power (dark blue color) used for training the model is available only when the house is added to the training set. For the test houses, we predict the appliance power denoted by a "?" mark and validate our performance.
• Figure 3: Seq2Point architecture of a) Single Output and b) Multi Output
• Figure 4: Two aggregate function choices: a) uniform, b) triangle
• Figure 5: Different aggregation time window choices: Dynamic $\mathrm{DP1}$ for $\mathrm{T1}$ and $\mathrm{DP2}$ for $\mathrm{T2}$ or any static window ($\mathrm{SP1, SP2, SP3, SP4}$).