Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

COLD: Concurrent Loads Disaggregator for Non-Intrusive Load Monitoring

Ilia Kamyshev, Sahar Moghimian Hoosh, Dmitrii Kriukov, Elena Gryazina, Henni Ouerdane

TL;DR

Problem: disaggregate high-frequency NILM data with many concurrent devices. Approach: COLD, a transformer-based, multi-output NILM model that processes spectrogram inputs and enforces energy conservation with a softmax head. Contributions: (i) state-of-the-art identification accuracy $F1=0.95$ and disaggregation $MF=0.82$ under up to 11 concurrent loads, (ii) a fully labeled UK-DALE 16 kHz NILM dataset with ${85k}$ training, ${5k}$ validation, and ${10k}$ test samples, and (iii) an analysis of performance degradation as concurrency rises. Significance: demonstrates scalable high-frequency NILM capable of handling many devices and provides data/code for reproducibility.

Abstract

The global effort toward renewable energy and the electrification of energy-intensive sectors have significantly increased the demand for electricity, making energy efficiency a critical focus. Non-intrusive load monitoring (NILM) enables detailed analyses of household electricity usage by disaggregating the total power consumption into individual appliance-level data. In this paper, we propose COLD (Concurrent Loads Disaggregator), a transformer-based model specifically designed to address the challenges of disaggregating high-frequency data with multiple simultaneously working devices. COLD supports up to 42 devices and accurately handles scenarios with up to 11 concurrent loads, achieving 95% load identification accuracy and 82% disaggregation performance on the test data. In addition, we introduce a new fully labeled high-frequency NILM dataset for load disaggregation derived from the UK-DALE 16 kHz dataset. Finally, we analyze the decline in NILM model performance as the number of concurrent loads increases.

COLD: Concurrent Loads Disaggregator for Non-Intrusive Load Monitoring

TL;DR

Problem: disaggregate high-frequency NILM data with many concurrent devices. Approach: COLD, a transformer-based, multi-output NILM model that processes spectrogram inputs and enforces energy conservation with a softmax head. Contributions: (i) state-of-the-art identification accuracy and disaggregation under up to 11 concurrent loads, (ii) a fully labeled UK-DALE 16 kHz NILM dataset with training, validation, and test samples, and (iii) an analysis of performance degradation as concurrency rises. Significance: demonstrates scalable high-frequency NILM capable of handling many devices and provides data/code for reproducibility.

Abstract

The global effort toward renewable energy and the electrification of energy-intensive sectors have significantly increased the demand for electricity, making energy efficiency a critical focus. Non-intrusive load monitoring (NILM) enables detailed analyses of household electricity usage by disaggregating the total power consumption into individual appliance-level data. In this paper, we propose COLD (Concurrent Loads Disaggregator), a transformer-based model specifically designed to address the challenges of disaggregating high-frequency data with multiple simultaneously working devices. COLD supports up to 42 devices and accurately handles scenarios with up to 11 concurrent loads, achieving 95% load identification accuracy and 82% disaggregation performance on the test data. In addition, we introduce a new fully labeled high-frequency NILM dataset for load disaggregation derived from the UK-DALE 16 kHz dataset. Finally, we analyze the decline in NILM model performance as the number of concurrent loads increases.

Paper Structure

This paper contains 13 sections, 7 equations, 7 figures, 1 table.

Table of Contents

  1. Introduction
  2. Our contributions
  3. Preparation of UK-DALE dataset
  4. Concurrent Loads Disaggregator (COLD)
  5. Features extraction
  6. Transformer encoder layers
  7. Multi-head self-attention
  8. Head of COLD
  9. Loss function
  10. Evaluation metrics
  11. Results
  12. Discussion
  13. Conclusion

Figures (7)

  • Figure 1: Overview of the NILM technology
  • Figure 2: Distribution of the number of simultaneously active devices in the House 1 from the UK-DALE dataset.
  • Figure 3: One of the steady-state segments extracted from the House 1 of the UK-DALE dataset (a) and its spectrogram (b).
  • Figure 4: COLD architecture layout.
  • Figure 5: F1 and MF score curves over 700 training epochs.
  • ...and 2 more figures