Short-term power load forecasting model based on dynamic convolution decomposition and multi-scale graph

doi:10.11772/j.issn.1001-9081.2025060676

Journal of Computer Applications ›› 2026, Vol. 46 ›› Issue (7): 2334-2346.DOI: 10.11772/j.issn.1001-9081.2025060676

• Frontier and comprehensive applications • Previous Articles

Short-term power load forecasting model based on dynamic convolution decomposition and multi-scale graph

Li ZHU¹, Lugema MI¹(), Chunqiang ZHU²^,³, Wanru XU¹, Jingkai GAO¹, Jinqi QU¹

^1.College of Artificial Intelligence and Computer Science，Xi'an University of Science and Technology，Xi'an Shaanxi 710600，China
^2.School of Computer Science and Technology，Xi'an Jiaotong University，Xi'an Shaanxi 710049，China
^3.Training Center，State Grid Shaanxi Electric Power Company Limited，Xi'an Shaanxi 710032，China

Received:2025-06-19 Revised:2025-09-13 Accepted:2025-09-29 Online:2025-10-13 Published:2026-07-10
Contact: Lugema MI
About author:ZHU Li， born in 1976， M. S.， associate professor. Her research interests include intelligent information processing.
ZHU Chunqiang， born in 1974， Ph. D. candidate， senior engineer. His research interests include intelligent analysis of power systems.
XU Wanru， born in 2000， M. S. candidate. Her research interests include time series data generation.
GAO Jingkai， born in 1996， M. S. candidate. His research interests include machine learning.
QU Jinqi， born in 2003. His research interests include artificial intelligence， software development.
Supported by:
National Natural Science Foundation of China(52474247);Science and Technology Project of State Grid Shaanxi Electric Power Company Limited(5226PX250001)

基于动态卷积分解和多尺度图的短期电力负荷预测模型

朱莉¹, 米路革麻¹(), 朱春强²^,³, 徐婉茹¹, 高靖凯¹, 屈锦琪¹

^1.西安科技大学人工智能与计算机学院，西安 710600
^2.西安交通大学计算机科学与技术学院，西安 710049
^3.国网陕西省电力公司培训中心，西安 710032

通讯作者: 米路革麻
作者简介:朱莉（1976—），女，陕西宝鸡人，副教授，硕士，CCF会员，主要研究方向：智能信息处理
朱春强（1974—），男，陕西汉中人，高级工程师，博士研究生，主要研究方向：电力系统智能分析
徐婉茹（2000—），女，陕西宝鸡人，硕士研究生，主要研究方向：扩散模型、时间序列数据生成
高靖凯（1996—），男，河南周口人，硕士研究生，主要研究方向：机器学习
屈锦琪（2003—），男，陕西宝鸡人，主要研究方向：人工智能、软件开发。
基金资助:
国家自然科学基金资助项目(52474247);国网陕西省电力有限公司科技项目(5226PX250001);国网陕西省电力有限公司科技项目(5226PX250003)

Abstract

Abstract:

To address the problem that the existing short-term power load forecasting methods have difficulty in effectively modeling non-linear structures and lack cross-scale and cross-variable interaction capabilities， a short-term power load forecasting model based on Dynamic Convolution Decomposition and Multi-Scale Graph （DCDMSG） was proposed. First，to cope with the complex non-linear structure， a dynamic convolution decomposition method was adopted to decompose the trend and seasonal items in the load sequence. Second， the trend item was forecasted using a linear layer directly， and for the seasonal item， a multi-scale sequence was constructed for deep modeling using an adaptive multi-scale sequence construction method. Third， to model the complex dependencies within and outside the multi-scale sequence， a multi-scale fusion graph was used to capture dependencies within sequences at different scales， a multi-variable correlation graph was utilized to model correlations between variables， and a multi-scale mix-hop propagation was employed to aggregate features in the graph. Finally， the seasonal items were forecasted， and the obtained forecast results were weighted and fused with the trend items' forecast results to obtain the final forecast values. Experimental results show that， on the Australian dataset， the Mean Absolute Error （MAE） and Root Mean Square Error （RMSE） of DCDMSG reached 0.379 MW and 0.529 MW， respectively， which were reduced by 7.11%-21.37% and 9.88%-26.32% compared with all comparison models. On the Cele dataset， DCDMSG achieved an MAE of 0.437 MW and an RMSE of 0.708 MW， representing decreases of 5.61%-24.66% and 5.22%-24.12% over all comparison models. In the short-term power load forecasting task， DCDMSG effectively improves the forecast accuracy through seasonal-trend decomposition as well as cross-scale and cross-variable modeling.

Key words: short-term power load forecasting, dynamic convolution decomposition, multi-scale feature, Graph Neural Network (GNN), mix-hop propagation

摘要：

针对现有的短期电力负荷预测方法难以有效建模非线性结构和缺乏跨尺度与跨变量交互能力的问题，提出基于动态卷积分解和多尺度图的短期电力负荷预测模型（DCDMSG）。首先，采用动态卷积分解方法分解负荷序列中的趋势和季节项，以应对复杂的非线性结构；其次，利用线性层预测趋势项，对季节项则采用自适应多尺度序列构建方法生成多尺度序列并进行深度建模；再次，为了建模多尺度序列内外的复杂依赖关系，使用多尺度融合图捕捉不同尺度序列内的依赖关系，利用多变量相关图建模不同变量之间的相关性，并采用多尺度混合跳传播聚合图中的特征；最后，对季节项进行预测，并将得到的预测结果与趋势项预测结果进行加权融合得到最终预测值。实验结果表明，在Australian数据集上， DCDMSG的平均绝对误差（MAE）为0.379 MW，均方根误差（RMSE）为0.529 MW，比对比模型分别降低了7.11%~21.37%和9.88%~26.32%；在Cele数据集上， DCDMSG的MAE为0.437 MW， RMSE为0.708 MW，比对比模型分别降低了5.61%~24.66%和5.22%~24.12%。在短期电力负荷预测任务中，DCDMSG利用季节趋势分解和跨尺度与跨变量建模有效提高了预测精度。

关键词: 短期负荷预测, 动态卷积分解, 多尺度特征, 图神经网络, 混合跳传播

CLC Number:

TP183

Li ZHU, Lugema MI, Chunqiang ZHU, Wanru XU, Jingkai GAO, Jinqi QU. Short-term power load forecasting model based on dynamic convolution decomposition and multi-scale graph[J]. Journal of Computer Applications, 2026, 46(7): 2334-2346.

朱莉, 米路革麻, 朱春强, 徐婉茹, 高靖凯, 屈锦琪. 基于动态卷积分解和多尺度图的短期电力负荷预测模型[J]. 《计算机应用》唯一官方网站, 2026, 46(7): 2334-2346.

Figures/Tables 22

References 29

[1]	杨佳泽，王灿，王增平.新型电力系统背景下的智能负荷预测算法研究综述［J］.华北电力大学学报（自然科学版）， 2025， 52（3）： 54-67.
	Yang Jiaze， Wang Can， Wang Zengping. Review on intelligent load forecasting algorithms for the new-type power system ［J］. Journal of North China Electric Power University （Natural Science Edition）， 2025， 52（3）： 54-67.
[2]	史含笑，王雷春.结合LSTM和自注意力机制的图卷积网络短期电力负荷预测［J］.计算机应用， 2024， 44（1）： 311-317.
	Shi Hanxiao， Wang Leichun. Short-term power load forecasting by graph convolutional network combining LSTM and self-attention mechanism ［J］. Journal of Computer Applications， 2024， 44（1）： 311-317.
[3]	Zhou S， Li Y， Guo Y， et al. A load forecasting framework considering hybrid ensemble deep learning with two-stage load decomposition ［J］. IEEE Transactions on Industry Applications， 2024， 60（3）： 4568-4582.
[4]	Li J， Deng D， Zhao J， et al. A novel hybrid short-term load forecasting method of smart grid using MLR and LSTM neural network ［J］. IEEE Transactions on Industrial Informatics， 2021， 17（4）： 2443-2452.
[5]	徐任超，阎威武，王国良，等.基于周期性建模的时间序列预测方法及电价预测研究［J］.自动化学报， 2020， 46（6）： 1136-1144.
	Xu Renchao， Yan Weiwu， Wang Guoliang， et al. Time series forecasting based on seasonality modeling and its application to electricity price forecasting ［J］. Acta Automatica Sinica， 2020， 46（6）： 1136-1144.
[6]	Peng H， Lou Y， Li F， et al. Decomposition framework for long term load forecasting on temperature insensitive area ［J］. Energy Reports， 2024， 12： 5783-5792.
[7]	Hu S， Wang Y， Cai W， et al. A combined method for short-term load forecasting considering the characteristics of components of seasonal and trend decomposition using local regression ［J］. Applied Sciences， 2024， 14（6）： No.2286.
[8]	Karamolegkos S， Koulouriotis D E. Advancing short-term load forecasting with decomposed Fourier ARIMA： a case study on the Greek energy market ［J］. Energy， 2025， 325： No.135854.
[9]	Wu H， Xu J， Wang J， et al. Autoformer： decomposition transformers with auto-correlation for long-term series forecasting ［C］// NeurIPS 2021. Red Hook： Curran Associates Inc.， 2021： 22419-22430.
[10]	Zhou T， Ma Z， Wen Q Q， et al. FEDformer： frequency enhanced decomposed transformer for long-term series forecasting ［C］// ICML 2022. New York： JMLR.org， 2022： 27268-27286.
[11]	Zeng A， Chen M， Zhang L， et al. Are Transformers effective for time series forecasting ［C］// AAAI 2023. Palo Alto： AAAI Press， 2023： 11121-11128.
[12]	Liu S， Yu H， Liao C， et al. Pyraformer： low-complexity pyramidal attention for long-range time series modeling and forecasting ［EB/OL］. （2023-02-14）［2025-08-26］. .
[13]	Zhou H， Zhang S， Peng J， et al. Informer： beyond efficient transformer for long sequence time-series forecasting ［C］// AAAI 2021. Palo Alto： AAAI Press， 2021： 11106-11115.
[14]	Ahmad M， Butt M H F， Mazzara M， et al. Pyramid hierarchical spatial-spectral transformer for hyperspectral image classification ［J］. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing， 2024， 17： 17681-17689.
[15]	Chen P， Zhang Y， Cheng Y， et al. Pathformer： multi-scale transformers with adaptive pathways for time series forecasting ［PP/OL］. V5. arXiv （2024-12-15）［2025-08-27］. .
[16]	Nie Y， Nguyen N H， Sinthong P， et al. A time series is worth 64 words： long-term forecasting with Transformers ［PP/OL］. V2. arXiv （2023-03-04）［2025-08-27］. .
[17]	Zhang Y， Yan J. Crossformer： Transformer utilizing cross-dimension dependency for multivariate time series forecasting ［EB/OL］. （2023-03-02）［2025-08-27］. .
[18]	Zhang Z， Meng L， Gu Y. SageFormer： series-aware framework for long-term multivariate time series forecasting ［J］. IEEE Internet of Things Journal， 2024， 11（10）： 18435-18448.
[19]	Jin M， Huan Y， Wen Q， et al. A survey on graph neural networks for time series： forecasting， classification， imputation， and anomaly detection ［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence， 2024， 46（12）： 10466-10485.
[20]	李晓寒，王俊，贾华丁，等.基于多重注意力机制的图神经网络股市波动预测方法［J］.计算机应用， 2022， 42（7）： 2265-2273.
	Li Xiaohan， Wang Jun， Jia Huading， et al. Stock market volatility prediction method based on graph neural network with multi-attention mechanism ［J］. Journal of Computer Applications， 2022， 42（7）： 2265-2273.
[21]	Cai W， Wang K， Wu H， et al. ForecastGrapher： redefining multivariate time series forecasting with graph neural networks ［PP/OL］. arXiv （2024-05-28）［2025-08-27］. .
[22]	Cai W， Liang Y， Feng J， et al. MSGNet： learning multi-scale inter-series correlations for multivariate time series forecasting ［C］// AAAI 2024. Palo Alto： AAAI Press， 2024： 11141-11149.
[23]	Sun C， Ning Y， Shen D， et al. Graph neural network-based short-term load forecasting with temporal convolution ［J］. Data Science and Engineering， 2024， 9（2）： 113-132.
[24]	Wu Z， Pan S， Long G， et al. Connecting the dots： multivariate time series forecasting with graph neural networks ［C］// KDD 2020. New York： ACM， 2020： 753-763.
[25]	Cao D， Wang Y， Duan J， et al. Spectral temporal graph neural network for multivariate time-series forecasting ［C］// NeurIPS 2020. Red Hook： Curran Associates Inc.， 2020： 17766-17778.
[26]	Yi K， Zhang Q， Fan W， et al. FourierGNN： rethinking multivariate time series forecasting from a pure graph perspective ［C］// NeurIPS 2023. Red Hook： Curran Associates Inc.， 2023： 69638-69660.
[27]	Zhao X， Yu W， Zhang Y. MST-GNN： graph neural network with multi-granularity in space and time for traffic prediction ［J］. GeoInformatica， 2025， 29（3）： 377-401.
[28]	Luo X， Wang H， Han T， et al. FFT-Trans： enhancing robustness in mechanical fault diagnosis with Fourier transform-based transformer under noisy conditions ［J］. IEEE Transactions on Instrumentation and Measurement， 2024， 73： No.2515112.
[29]	Liang A， Chai X， Sun Y， et al. GTformer： graph-based temporal-order-aware Transformer for long-term series forecasting ［J］. IEEE Internet of Things Journal， 2024， 11（19）： 31467-31478.

超参数	设置
Batch size	64
Dimension of model	512
Learning rate	0.000 1
Dropout	0.05
Epochs	25
GCN depth	5
Activation function	GELU
Loss function	MSE

超参数	设置
Batch size	64
Dimension of model	512
Learning rate	0.000 1
Dropout	0.05
Epochs	25
GCN depth	5
Activation function	GELU
Loss function	MSE

模型	MAE/MW	RMSE/MW	MAPE/%
DCDMSG	0.379	0.529	2.643
PatchTST	0.409	0.587	2.872
FEDformer	0.471	0.633	3.643
Autoformer	0.451	0.624	3.243
MTGNN	0.482	0.718	3.756
FourierGNN	0.408	0.692	3.146

模型	MAE/MW	RMSE/MW	MAPE/%
DCDMSG	0.379	0.529	2.643
PatchTST	0.409	0.587	2.872
FEDformer	0.471	0.633	3.643
Autoformer	0.451	0.624	3.243
MTGNN	0.482	0.718	3.756
FourierGNN	0.408	0.692	3.146

模型	MAE/MW	RMSE/MW	MAPE/%
DCDMSG	0.437	0.708	2.774
PatchTST	0.463	0.751	2.939
FEDformer	0.496	0.821	3.151
Autoformer	0.506	0.761	3.397
MTGNN	0.580	0.933	3.851
FourierGNN	0.491	0.747	2.901

Short-term power load forecasting model based on dynamic convolution decomposition and multi-scale graph

基于动态卷积分解和多尺度图的短期电力负荷预测模型

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 22

References 29

Related Articles 15

Recommended Articles

Metrics

卷积核组合	MAE/MW	RMSE/MW	MAPE/%
3，7，15，25	0.379	0.529	2.643
15，25	0.391	0.571	2.912
3，7	0.420	0.593	3.099
30	0.415	0.573	2.937
15	0.419	0.577	2.931
7	0.423	0.596	3.236

模型	MAE/MW	RMSE/MW	MAPE/%
多尺度+多变量	0.379	0.529	2.643
多尺度+单变量	0.426	0.613	3.324
单尺度+多变量	0.428	0.607	3.195
单尺度+单变量	0.438	0.694	4.031

噪声环境	场景	σ	µ
弱噪声	1	0.05×std	0.05
弱噪声	2	0.05×std	0.10
中噪声	3	0.10×std	0.10
中噪声	4	0.10×std	0.20
强噪声	5	0.20×std	0.20
强噪声	6	0.20×std	0.30

噪声环境	场景	DCDMSG		PatchTST		FourierGNN
噪声环境	场景	MAE/MW	RMSE/MW	MAE/MW	RMSE/MW	MAE/MW	RMSE/MW
弱	1	0.469	0.621	0.517	0.690	0.511	0.801
弱	2	0.596	0.775	0.645	0.865	0.637	0.970
中	3	0.594	0.773	0.644	0.866	0.636	0.970
中	4	0.749	1.055	0.784	1.116	0.780	1.221
强	5	0.749	1.050	0.782	1.122	0.781	1.225
强	6	0.804	1.160	0.837	1.218	0.828	1.324

模型	时间复杂度	空间复杂度
DCDMSG	O（T）	O（T）
PatchTST	O（T²）	O（T²）
FEDformer	O（T）	O（T）
Autoformer	O（T log T）	O（T log T）
MTGNN	O（N²）	O（N²）
FourierGNN	O（N log N）	O（N²）

[1]	Xiangyi WU, Hailiang YE, Feilong CAO. Point cloud completion method based on smooth-sharpen graph convolution [J]. Journal of Computer Applications, 2026, 46(7): 2267-2276.
[2]	Yue ZHANG, Ying WANG. Scalable dynamic graph representation learning with neighborhood memory [J]. Journal of Computer Applications, 2026, 46(7): 2104-2110.
[3]	Min LU, Hui ZHOU. Airport gate assignment algorithm based on node prediction in conflict graph of assigned activities [J]. Journal of Computer Applications, 2026, 46(6): 1913-1921.
[4]	Chao LYU, Geyao MA. Lightweight human pose estimation network based on redundant feature suppression [J]. Journal of Computer Applications, 2026, 46(6): 1973-1980.
[5]	Hang QI, Tingting DONG, Yongqiang NAI, Xian MO. Contrastive collaborative filtering method based on graph diffusion generation and adaptive sampling [J]. Journal of Computer Applications, 2026, 46(6): 1818-1828.
[6]	Hongrui ZHANG, Weiming FENG, Luxia YANG, Yongjie MA. CSAF-YOLO： improved YOLO11 algorithm for underwater small object detection [J]. Journal of Computer Applications, 2026, 46(5): 1578-1585.
[7]	Minqi WU, Yuanhua YANG, Hang LI, Yaqin HU, Zhihao TANG, Teng MEI. Lightweight underwater small object detection based on graph Transformer and RT-DETR [J]. Journal of Computer Applications, 2026, 46(5): 1586-1595.
[8]	Kun FU, Haoyu WEI, Weijing LIU, Xing DANG, Zezheng LIU, Jianwei LI. Graph neural network framework for topology semantic dual-domain collaboration [J]. Journal of Computer Applications, 2026, 46(5): 1378-1387.
[9]	Fengwei CHENG, Bingqi ZHANG, Guohua XU, Wenjian WANG. Competitive loss-driven generative imbalanced node classification [J]. Journal of Computer Applications, 2026, 46(5): 1475-1481.
[10]	Xumeng DOU, Bin XIE, Zhaohui ZHANG, Zhengang ZHAO, Hanyu DUAN, Aolei GUO. Drug-target interaction prediction based on structure-network collaborative features and grid-attention enhanced Kolmogorov-Arnold network [J]. Journal of Computer Applications, 2026, 46(4): 1344-1353.
[11]	Xinyi YAN, Linglong ZHU, Yonghong ZHANG. CDC-DETR： multi-scale real-time human-vehicle detection method for complex traffic scenarios [J]. Journal of Computer Applications, 2026, 46(4): 1283-1291.
[12]	Hanqing LIU, Guoming SANG, Yijia ZHANG. Remote sensing image captioning model combining dense multi-scale feature fusion and feature knowledge-enhanced Transformer [J]. Journal of Computer Applications, 2026, 46(3): 741-749.
[13]	Rilong WANG, Zhenping LI, Xiaosong LI, Qiang GAO, Ya HE, Yong ZHONG, Yingxiao ZHAO. Multi-Agent collaborative knowledge reasoning framework [J]. Journal of Computer Applications, 2026, 46(3): 708-714.
[14]	Chunyong YIN, Bufan ZHANG. Multi-scale based multivariate time series anomaly detection model [J]. Journal of Computer Applications, 2026, 46(3): 790-797.
[15]	Jinjiao LIN, Canshun ZHANG, Shuya CHEN, Tianxin WANG, Jian LIAN, Yonghui XU. Vehicle insurance fraud detection method based on improved graph attention network [J]. Journal of Computer Applications, 2026, 46(2): 437-444.

模型	MAE/MW	RMSE/MW	MAPE/%
ALL	0.379	0.529	2.643
Only DCDF	0.431	0.732	4.032
Only MSG	0.408	0.697	3.776
Only Mmixhop	0.450	0.795	4.553
Baseline	0.469	1.353	5.637

模型	MAE/MW	RMSE/MW	MAPE/%
ALL	0.379	0.529	2.643
Only DCDF	0.431	0.732	4.032
Only MSG	0.408	0.697	3.776
Only Mmixhop	0.450	0.795	4.553
Baseline	0.469	1.353	5.637