.png){kind=icon}
.png){kind=icon}
Return
Dynamic and Memory-Efficient Shape-Based Methodologies for User Type Identification in Smart Grid Applications
.png){kind=icon}
PRE
AI
DOI：10.1109/TII.2025.3625552.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Behind-the-meter (BTM) equipment identification and monitoring their availability in real time in smart grid applications require computationally efficient methodologies suitable for edge deployment. Existing shape-based approaches, whilst maintaining interpretability advantages over structure-based methods, suffer from computational bottlenecks and memory constraints when processing streaming data. This article develops three dynamic and memory-efficient updating strategies for similarity profile computation: an additive method for lossless updates, a fixed-memory approach with configurable inertia strategies, and a codebook-based technique employing dictionary learning for compressed data representation. The proposed methodologies eliminate the requirement for complete historical data reprocessing during each update, addressing critical limitations in edge computing environments. Comprehensive simulation studies using real-world photovoltaic user data demonstrate that the codebook-based approach achieves over 30% memory reduction while maintaining classification accuracy. The fixed-memory technique exhibits superior performance for applications requiring rapid change detection, with different inertia strategies providing varying sensitivity levels for diverse smart grid applications. These dynamic methodologies enable the practical deployment of interpretable BTM identification systems on resource-constrained edge devices whilst preserving the pattern recognition advantages of shape-based approaches.
Keywords:
Binary classification
data compression
data mining
dynamic updating
pattern recognition
renewable energy
time-series mining
Journal
IF:
9.9
Papers:
8.3K
Citations:
6.0W