A load forecasting method for integrated energy systems based on depthwise separable temporal convolution and graph convolution

WANG Jidong; CHAI Yihan; ZHOU Yue; KONG Xiangyu

doi:10.19783/j.cnki.pspc.260039

English Version Journal Login

您当前的位置：

首页 >

文章列表页 >

A load forecasting method for integrated energy systems based on depthwise separable temporal convolution and graph convolution

更新时间：2026-07-19

- A load forecasting method for integrated energy systems based on depthwise separable temporal convolution and graph convolution
- Power System Protection and Control Vol. 54, Issue 14, (2026)
- 作者机构：
- 作者简介：
- 基金信息：
- DOI：10.19783/j.cnki.pspc.260039
  CLC：
- Published：2026
- 稿件说明：
移动端阅览
WANG Jidong, CHAI Yihan, ZHOU Yue, et al. A load forecasting method for integrated energy systems based on depthwise separable temporal convolution and graph convolution[J]. Power System Protection and Control, 2026, 54(14).
DOI：

WANG Jidong, CHAI Yihan, ZHOU Yue, et al. A load forecasting method for integrated energy systems based on depthwise separable temporal convolution and graph convolution[J]. Power System Protection and Control, 2026, 54(14). DOI： 10.19783/j.cnki.pspc.260039.

摘要

在综合能源系统中，同一区域的多能负荷既存在跨能量耦合关系，又存在跨节点的时空依赖关系。针对传统独立建模难以刻画跨能量、跨节点依赖的问题，提出一种融合深度可分离时间卷积网络和图卷积网络的负荷预测方法。首先，根据用能特征相似性筛选用能单元，构建负荷数据集。然后，构建深度可分离时间卷积网络与图卷积网络的融合框架，并引入多头稀疏注意力模块以捕获关键时刻交互信息。最后，采用自适应门控机制融合深度可分离时间卷积网络和图卷积网络。实验结果表明，所提方法在综合能源系统的负荷预测中具有更高的精度，误差小于对比模型，并在单步与多步预测任务上均展现出有效性。

Abstract

In integrated energy systems

multi-energy loads within the same region exhibit not only cross-energy coupling relationships but also cross-node spatiotemporal dependencies. To address the difficulty of capturing these cross-energy and cross-node dependencies using traditional independent modeling approaches

a load forecasting method that integrates a deep separable temporal convolutional network with a graph convolutional network is proposed. First

energy-consuming units are selected based on similarities of their energy consumption characteristics to construct the load dataset. Next

a fusion framework combining a deep separable temporal convolutional network and a graph convolutional network is established

incorporating a multi-head sparse attention module to capture key temporal interaction information at critical time steps. Finally

an adaptive gating mechanism is employed to integrate the deep separable temporal convolutional network and the graph convolutional network. Experimental results demonstrate that the proposed method achieves higher accuracy in load forecasting for integrated energy systems

with lower prediction errors

and exhibits superior performance in both single-step and multi-step forecasting tasks.

关键词

Keywords

references

ZHU X, SHEN M. Based on the ARIMA model with grey theory for short term load forecasting model[C] // 2012 International Conference on Systems and Informatics (ICSAI2012), Yantai, China, 2012: 564-567.

LIU L, BAI F, SU C, et al. Forecasting the occurrence of extreme electricity prices using a multivariate logistic regression model[J]. Energy, 2022, 247.

ZHU S, MA H, CHEN L, et al. Short-term load forecasting of an integrated energy system based on STL-CPLE with multitask learning[J]. Protection and Control of Modern Power Systems, 2024, 9(6): 71-92.

LI Z, YE L, SONG X, et al. Heterogeneous spatiotemporal graph convolution network for multi-modal wind-PV power collaborative prediction[J]. IEEE Transactions on Power Systems, 2024, 39(4): 5591-5608.

YANG Y, LIU Y, ZHANG Y, et al. DEST-GNN: a double-explored spatio-temporal graph neural network for multi-site intra-hour PV power forecasting[J]. Applied Energy, 2025, 378.

LI Y, LI C, CHEN G, et al. Multi-task graph adaptive learning for multivariate electricity price short-term forecasting in Australia’s national electricity market[J]. IEEE Transactions on Power Systems, 2025, 40(1): 530-542.

ZHUANG W, FAN J, XIA M, et al. A multi-scale spatial-temporal graph neural network-based method of multienergy load forecasting in integrated energy system[J]. IEEE Transactions on Smart Grid, 2024, 15(3): 2652-2666.

PENG D, LIU Y, WANG D, et al. Multi-energy load forecasting for integrated energy system based on sequence decomposition fusion and factors correlation analysis[J]. Energy, 2024, 308.

LI M, WANG W, HE Y, et al. Deep learning model for short-term photovoltaic power forecasting based on variational mode decomposition and similar day clustering[J]. Computers and Electrical Engineering, 2024, 115: 109-116.

ZHENG C, FAN X, WANG C, et al. GMAN: a graph multi-attention network for traffic prediction[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(1): 1234-1241.

National Renewable Energy Laboratory. Commercial and residential hourly load profiles for all TMY3 locations in the United States[Z]. 2014.

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Short-term power load forecasting based on multi-scale parallel feature fusion and Kolmogorov-Arnold network reconstruction

Optimal coordinated attack strategy for distributed economic dispatch of electricity-heat-gas integrated energy systems

Whole life-cycle zero-carbon planning and multi-time-scale energy optimization for integrated energy systems

Optimal scheduling of integrated energy systems based on prior knowledge and multi-stage QMIX reinforcement learning

基于深度强化学习的区域综合能源系统主动调节灵活性规则提取及可解释优化调度

Related Author

YU Yongjin

JIAN Yilin

LIU Qi

JIAO Wenshu

CHENG Shan

WAN Siyu

LIU Weiwei

MA Bingtai

Related Institution

College of Electrical Engineering and Automation, Shandong University of Science and Technology

Hubei Engineering Research Center of New Energy Microgrid (China Three Gorges University)

中国长江三峡集团有限公司

三峡国际能源投资集团有限公司

中国电力科学研究院有限公司

AI问答

⁰