@article{Liu2026, 
author = {Weicheng Liu and Jianfeng Zhao and Yue Yu and Xiao-Ping Zhang and Ying Xue},
title = {Temporal Convolutional Auto-encoder for Residential Non-intrusive Load Monitoring},
year = {2026},
journal = {CSEE Journal of Power and Energy Systems},
volume = {12},
number = {3},
pages = {1208-1220},
keywords = {Demand-side response, dilated causal convolution, load monitoring, supervised learning},
url = {https://www.sciopen.com/article/10.17775/CSEEJPES.2022.03070},
doi = {10.17775/CSEEJPES.2022.03070},
abstract = {Non-intrusive load monitoring (NILM) enables disaggregation of appliance power usage through data gathered from the incoming power line. This disaggregation assists householders in reducing electricity expenses and aids system operators in implementing demand-side responses. However, despite advancements in power measurement, processing household power data remains a significant challenge. Existing methods struggle to accurately disaggregate appliances with brief operation times and exhibit large errors in load curve regression tasks. To address these issues, a novel time-power hybrid (TPH) algorithm is proposed to enhance the accuracy of dataset preprocessing for NILM model training. Additionally, the temporal convolutional auto-encoder model (TCAE) is introduced to improve disaggregation performance. Both the TPH and TCAE are evaluated using a public dataset in comparison to other representative methods. The results demonstrate that the proposed TPH algorithm exhibits superior load activation extraction performance. The TCAE model displays significant advantages in both curve regression and load ON/OFF detection for various appliances. Remarkably, the TCAE achieves a detection precision of 99.9%, particularly for appliances with short operation times, which other models are unable to accomplish. Furthermore, the TCAE necessitates less computational space compared to existing models.}
}