Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Bayesian Optimization-Based AutoML Framework for Non-Intrusive Load Monitoring

Nazanin Siavash, Armin Moin

TL;DR

This work tackles NILM by embedding Automated Machine Learning within a Bayesian optimization framework, enabling automated model selection and hyperparameter tuning for energy disaggregation. The proposed AutoML4NILM toolkit supports 11 algorithms and offers a flexible, extensible search space, with Hyperopt-based SMBO guiding the search. Empirical results on the UK-DALE dataset identify Seq2Point as the strongest performer (MAE ~7.12), while highlighting limitations due to computational constraints on exhaustive hyperparameter exploration. Overall, AutoML4NILM delivers a practical, open-source pathway to apply advanced NILM techniques without deep ML expertise, accelerating research and deployment in energy disaggregation.

Abstract

Non-Intrusive Load Monitoring (NILM), commonly known as energy disaggregation, aims to estimate the power consumption of individual appliances by analyzing a home's total electricity usage. This method provides a cost-effective alternative to installing dedicated smart meters for each appliance. In this paper, we introduce a novel framework that incorporates Automated Machine Learning (AutoML) into the NILM domain, utilizing Bayesian Optimization for automated model selection and hyperparameter tuning. This framework empowers domain practitioners to effectively apply machine learning techniques without requiring advanced expertise in data science or machine learning. To support further research and industry adoption, we present AutoML4NILM, a flexible and extensible open-source toolkit designed to streamline the deployment of AutoML solutions for energy disaggregation. Currently, this framework supports 11 algorithms, each with different hyperparameters; however, its flexible design allows for the extension of both the algorithms and their hyperparameters.

A Bayesian Optimization-Based AutoML Framework for Non-Intrusive Load Monitoring

TL;DR

This work tackles NILM by embedding Automated Machine Learning within a Bayesian optimization framework, enabling automated model selection and hyperparameter tuning for energy disaggregation. The proposed AutoML4NILM toolkit supports 11 algorithms and offers a flexible, extensible search space, with Hyperopt-based SMBO guiding the search. Empirical results on the UK-DALE dataset identify Seq2Point as the strongest performer (MAE ~7.12), while highlighting limitations due to computational constraints on exhaustive hyperparameter exploration. Overall, AutoML4NILM delivers a practical, open-source pathway to apply advanced NILM techniques without deep ML expertise, accelerating research and deployment in energy disaggregation.

Abstract

Non-Intrusive Load Monitoring (NILM), commonly known as energy disaggregation, aims to estimate the power consumption of individual appliances by analyzing a home's total electricity usage. This method provides a cost-effective alternative to installing dedicated smart meters for each appliance. In this paper, we introduce a novel framework that incorporates Automated Machine Learning (AutoML) into the NILM domain, utilizing Bayesian Optimization for automated model selection and hyperparameter tuning. This framework empowers domain practitioners to effectively apply machine learning techniques without requiring advanced expertise in data science or machine learning. To support further research and industry adoption, we present AutoML4NILM, a flexible and extensible open-source toolkit designed to streamline the deployment of AutoML solutions for energy disaggregation. Currently, this framework supports 11 algorithms, each with different hyperparameters; however, its flexible design allows for the extension of both the algorithms and their hyperparameters.
Paper Structure (12 sections, 4 equations, 7 figures, 3 tables)

This paper contains 12 sections, 4 equations, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Literature review
  4. Method
  5. AutoML for NILM
  6. Bayesian Optimization for Optimal Model Selection and Hyperparameter Tuning
  7. Experimental setup
  8. Mean Absolute Error (MAE)
  9. Classification Accuracy
  10. Experimental results
  11. Conclusion and Future Works
  12. Acknowledgment

Figures (7)

  • Figure 1: Framework Overview of NILM
  • Figure 2: A comprehensive view of the presented work
  • Figure 3: Three Main Components for Bayesian Optimization
  • Figure 4: Classification Accuracy vs. MAE
  • Figure 5: Classification Accuracy for each Model
  • ...and 2 more figures