.png){kind=icon}
.png){kind=icon}
Return
Variance Transformation Principle and Contrastive Learning Framework-Based Demand Response Baseline Load Estimation Method
.png){kind=icon}
PRE
AI
DOI：10.1109/TSG.2026.3651644.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Customer baseline load (CBL) estimation holds the key to the successful implementation of demand response (DR) programs. In traditional control group methods, non-DR users are clustered into sub-clusters. DR users are matched with those in sub-clusters showing similar load patterns, and the matched users’ load data is used to estimate DR users’ CBL. In general, the existing methods can generate relatively accurate CBL estimation results as long as the number of matched users is adequate. However, the current control group methods suffer from two major limitations, which not only undermine the estimation accuracy but may also render the methods entirely ineffective: 1) these methods are highly vulnerable to uneven distribution of sub-clusters; 2) load data of the non-DR participating users is under-exploited. To this end, this paper proposes a CBL estimation method based on variance transformation principle and contrastive learning framework. Firstly, the variance transformation principle is revealed, which is the basis to generate enough samples showing correlation with the DR users. Secondly, a neural network model based on contrastive learning framework is established to further recognize the non-DR participating users showing highly similar load pattern to the DR users and extract the high-dimensional load features from the DR users and the matched users. Finally, the CBL estimation results can be obtained by the feature similarity-weight calculation. The effectiveness and robustness of the proposed method is validated by using multi-source datasets and comprehensive evaluation indicators.
Keywords:
Baseline load estimation
demand response
control group method
contrastive learning
feature extraction
Journal
IF:
9.8
Papers:
5.7K
Citations:
4.3W