Editorial
An Integrated Digital-Twin-Assisted Edge Collaboration and LSTM-GAT Model for Short-Term Power Load Forecasting in Industrial Parks
@ARTICLE{10.4108/eetsis.14155, author={Ruilin Li and Honghong Li and Yoke Seng Wong and Fengyang Long and Donghua Jiang and Jianhua Wu}, title={An Integrated Digital-Twin-Assisted Edge Collaboration and LSTM-GAT Model for Short-Term Power Load Forecasting in Industrial Parks}, journal={EAI Endorsed Transactions on Scalable Information Systems}, volume={13}, number={7}, publisher={EAI}, journal_a={SIS}, year={2026}, month={11}, keywords={Industrial Internet of Things, Digital Twin, Edge Computing, Industrial Park Distribution, Short-Term Power Load Forecasting, Energy Management, LSTM-GAT}, doi={10.4108/eetsis.14155} }- Ruilin Li
Honghong Li
Yoke Seng Wong
Fengyang Long
Donghua Jiang
Jianhua Wu
Year: 2026
An Integrated Digital-Twin-Assisted Edge Collaboration and LSTM-GAT Model for Short-Term Power Load Forecasting in Industrial Parks
SIS
EAI
DOI: 10.4108/eetsis.14155
Abstract
INTRODUCTION: Manufacturing-oriented industrial parks are increasingly characterised by flexible production lines, energy-intensive equipment, and highly coupled production–energy processes. In such smart manufacturing environments, accurate short-term power load forecasting is essential for coordinating production scheduling, equipment operation, and low-carbon energy management. However, existing forecasting methods often overlook edge-side training efficiency and spatial correlations among multi-node power loads. OBJECTIVES: This paper aims to develop an efficient and accurate short-term power load forecasting framework for Industrial Internet of Things-enabled industrial parks by integrating digital-twin-assisted edge collaborative training with temporal–spatial load modelling. METHODS: An integrated digital-twin-assisted mobile edge computing–Industrial Internet of Things framework and an LSTM-GAT forecasting model, named L-Ga, are proposed. The digital twin layer mirrors the communication, computing, energy, and data-quality states of edge nodes to support node selection and collaborative model training. The L-Ga model combines long short-term memory networks and graph attention networks to capture temporal evolution and spatial coupling among park-level, feeder-level, utility-system, office-area, and key-equipment loads. RESULTS: The digital-twin-assisted node-selection strategy reduced the weighted training cost by 21.63% compared with random node selection. The proposed L-Ga model achieved average mean absolute error, root mean square error, and mean absolute percentage error values of 2.29, 3.12, and 1.98%, respectively, outperforming the representative baseline models. CONCLUSION: The proposed framework improves both edge-side collaborative training efficiency and short-term power load forecasting accuracy. It provides an effective solution for production–energy coordination, intelligent power distribution, and low-carbon energy management in Industrial Internet of Things-enabled manufacturing parks.
Copyright © 2026 Ruilin Li et al., licensed to EAI. This is an open access article distributed under the terms of the CC BY-NC-SA 4.0, which permits copying, redistributing, remixing, transformation, and building upon the material in any medium so long as the original work is properly cited.