Resilient Event Detection Algorithm for Non-Intrusive Load Monitoring Under Non-Ideal Conditions Using Reinforcement Learning

Event detection is critical in a non-intrusive load monitoring (NILM) solution. NILM is essential for the implementation of some demand-side management (DSM) techniques. This article proposes an event detection algorithm based on reinforcement learning for NILM purposes (RLNILM). The proposed method...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on industry applications 2024-03, Vol.60 (2), p.2085-2094
Main Authors: Etezadifar, Mozaffar, Karimi, Houshang, Aghdam, Amir G., Mahseredjian, Jean
Format: Article
Language:English
Subjects:
Algorithms
Cybersecurity
Event detection
federated learning
Feedback
Heuristic algorithms
Load monitoring
Machine learning
Machine learning algorithms
Monitoring
NILM
Non-intrusive load monitoring
Performance enhancement
Power grids
Real-time systems
reinforcement learning
smart grid
Training
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Event detection is critical in a non-intrusive load monitoring (NILM) solution. NILM is essential for the implementation of some demand-side management (DSM) techniques. This article proposes an event detection algorithm based on reinforcement learning for NILM purposes (RLNILM). The proposed method employs a number of simpler traditional event detection algorithms, e.g., LLR voting, or SWDC, to train the RLNILM agent through a feedback system that separates the RLNILM agent from directly accessing consumers' data. The performance of the proposed RLNILM method is validated using the real-world data from the iAWE dataset under ideal and non-ideal conditions. Four test scenario groups include cases where the frequency, input electric signals, or access to crucial grid information varies significantly. In all scenarios, the RLNILM agent outperforms traditional event detection algorithms used in the feedback system. The results show not only the proposed architecture increases the cyber security of the customers connected to NILM services, but also it improves the performance of real-time event detection algorithms in non-ideal grid conditions.
ISSN:0093-9994
1939-9367
DOI:10.1109/TIA.2023.3307347