Related Experiment Video
Updated: Aug 5, 2026
03:31
End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
Published on: December 15, 2023
GridCL for fine-grained load profiling in smart grids under limited labels
Ling Zhang1, Jia Wang1, Wenhua Zhang1
1State Grid Sichuan Electric Power Corporation, Chengdu, China.
Plos One
|July 28, 2026
Summary
GridCL, a new self-supervised learning framework, effectively addresses the challenge of limited labeled data for smart grid load profiling. It enables accurate energy management and demand response with minimal labeled user data.
Area of Science:
- Smart Grid Technology
- Machine Learning
- Data Science
Background:
- Fine-grained load profiling is crucial for smart grid demand response and energy management.
- Supervised methods are hindered by a lack of high-quality labeled datasets.
Purpose of the Study:
- To introduce GridCL, a self-supervised contrastive learning framework for low-label load profiling in smart grids.
- To overcome data scarcity limitations in supervised load profiling.
Main Methods:
- GridCL utilizes paired daily-load views generated through conservative input perturbations.
- Perturbations include small temporal rolling, multiplicative perturbation, and energy renormalization.
- A temporal convolutional encoder learns discriminative representations from unlabeled data.
Main Results:
- GridCL demonstrated strong clustering quality on the AllCities benchmark (ARI: 0.648, NMI: 0.719, Silhouette: 0.620).
- Achieved a best city-level ARI of 0.804.
- With only 10% labeled users, GridCL reached 84.5% accuracy, remaining stable with increased labels.
Conclusions:
- GridCL offers an effective low-label solution for fine-grained load profiling.
- The framework is suitable for practical smart-grid applications requiring efficient energy management.
