@article{SHANG2026, 
author = {Liqun SHANG and Danming JIA and Shuping GAO and Di AN and Xiaoqi LIU},
title = {A digital twin load forecasting model based on SARIMA-HKELM},
year = {2026},
journal = {Electric Power Engineering Technology},
volume = {45},
number = {10},
pages = {128-136},
keywords = {short-term electric power load forecasting, digital twin, seasonal autoregressive integrated moving average (SARIMA) model, hybrid kernel extreme learning machine (HKELM), hierarchical density-based spatial clustering of applications with noise (HDBSCAN), improved particle swarm optimization (IPSO) algorithm},
url = {https://www.sciopen.com/article/10.12158/j.2096-3203.2026.10.013},
doi = {10.12158/j.2096-3203.2026.10.013},
abstract = {Aiming at the problems that traditional neural networks have long prediction time and are difficult to update autonomously in load forecasting, which cannot meet the requirements of smart grids, a digital twin based seasonal autoregressive integrated moving average (SARIMA) model is proposed. Through distrust feature analysis, the load predict digital twin (LP-DT) model is replaced by an improved particle swarm optimization-hybrid kernel extreme learning machine (IPSO-HKELM) algorithm. Firstly, comprehensive similarity analysis is conducted on points with poor prediction performance in historical data, and the hierarchical density-based spatial clustering of applications with noise (HDBSCAN) algorithm is adopted to obtain the distrust point set and distrust features. Then, on the basis of the distrust feature point set, the IPSO algorithm is used to optimize hyperparameters such as kernel weights in the hybrid kernel extreme learning machine, and load forecasting is performed after training with historical distrust points. Finally, case analysis and comparison with traditional models are carried out, with the goodness-of-fit coefficient, root mean square error, mean absolute error and mean absolute percentage error as evaluation indicators. The proposed model achieves the values of 0.983, 53.87 kW, 40.55 kW and 0.74% respectively, which are significantly better than those of other models. This verifies the effectiveness and superiority of the proposed method in short-term load forecasting, and provides a reference for the design of load forecasting systems in smart grids.}
}