基于自适应时空图神经网络的居民区净负荷预测

doi:10.16182/j.issn1004731x.joss.25-0774

摘要/Abstract

摘要：

为解决传统的居民区整体用电负荷预测方法未充分考虑用户用电习惯的差异性而导致预测精度难以提升，提出了一种基于表示学习聚类与AGCN-Transformer相结合的居民区净负荷预测方法。利用表示学习方法充分提取用户净负荷数据的潜在特征，并基于用电行为的相似性将用户划分为不同的群体；对同一群体的净负荷数据进行聚合，并综合考虑聚类数目、用户历史净负荷序列以及不同群体间的相关性，构建适用于该任务的图结构；采用AGCN-Transformer模型对不同类别用户的未来负荷进行预测。实验结果表明：所提方法在预测精度与泛化能力方面均优于多种主流预测模型，展现出良好的应用前景。

关键词: 居民区, 表示学习聚类, 自适应图神经网络, Transformer, 净负荷预测

Abstract:

To address the issue that traditional residential overall power load forecasting methods fail to fully consider the differences in users' electricity consumption habits, making it difficult to improve prediction accuracy, a residential net load forecasting method combining representation learning clustering and an AGCN-Transformer was proposed. The representation learning method was utilized to fully extract the latent features of users' net load data, and users were divided into different groups based on the similarity of electricity consumption behaviors; the net load data of the same group were aggregated, and a graph structure suitable for this task was constructed by comprehensively considering the number of clusters, users' historical net load sequences, and the correlations among different groups; the AGCN-Transformer model was adopted to predict the future loads for users of different categories. Experimental results indicate that the proposed method outperforms multiple mainstream forecasting models in both prediction accuracy and generalization ability, showing a good application prospect.

Key words: residential area, representation learning clustering, adaptive graph neural network, Transformer, net load forecasting

中图分类号:

TP391.9

杨永琪,王聪,张宏立等 . 基于自适应时空图神经网络的居民区净负荷预测[J]. 系统仿真学报, 2026, 38(7): 2007-2019.

Yang Yongqi,Wang Cong,Zhang Hongli,et al . Adaptive Spatiotemporal Graph Neural Network-based Net Load Forecasting for Residential Areas[J]. Journal of System Simulation, 2026, 38(7): 2007-2019.

图/表 11

图1

图2

图3

表1

各季节聚类指标

月份	数据聚类簇数	轮廓系数	$I D B$	$V C H$
1	K=3	0.59	0.56	2 658.24
	K=4	0.61	0.51	2 871.79
	K=5	0.47	0.78	2 125.22
	K=6	0.70	0.47	3 767.02
	K=7	0.58	0.55	2 007.23
	K=8	0.53	1.52	992.43
4	K=3	0.60	0.53	2 524.86
	K=4	0.64	0.49	4 771.22
	K=5	0.49	0.56	2 056.05
	K=6	0.56	0.36	3 115.48
	K=7	0.54	0.67	3 663.21
	K=8	0.61	0.56	3 361.29
7	K=3	0.58	0.47	2 510.72
	K=4	0.47	0.49	2 415.01
	K=5	0.49	0.70	2 280.17
	K=6	0.63	0.54	4 303.38
	K=7	0.56	0.62	3 106.57
	K=8	0.47	0.83	2 161.75
10	K=3	0.60	0.52	1 443.49
	K=4	0.50	0.98	1 486.75
	K=5	0.48	0.59	1 700.7
	K=6	0.63	0.49	2 064.07
	K=7	0.59	0.56	2 044.82
	K=8	0.56	0.82	1 767.36

表1

图4

表2

表3

各季节代表月份预测指标

月份	方法	RSE/%	MAE/kW	RMSE/kW	$R C O R R$ /%
1	无聚类	8.34	66.66	82.73	98.12
	GCN	5.06	50.13	64.64	97.25
	Transformer	7.56	56.17	78.81	96.63
	GCN-Transformer	2.36	36.07	41.08	98.85
	AGCN-Transformer	1.72	30.64	37.69	99.36
4	无聚类	6.53	52.13	63.34	97.20
	GCN	5.12	43.98	56.08	97.48
	Transformer	7.08	51.02	65.95	97.41
	GCN-Transformer	7.93	57.21	71.94	96.91
	AGCN-Transformer	3.13	35.71	45.21	98.43
7	无聚类	13.14	61.65	82.98	96.01
	GCN	14.78	70.21	88.02	96.52
	Transformer	9.06	52.09	68.91	96.63
	GCN-Transformer	7.62	48.67	66.10	96.92
	AGCN-Transformer	5.29	41.00	52.66	97.48
10	无聚类	5.97	41.96	51.27	98.53
	GCN	15.84	95.32	106.65	95.62
	Transformer	5.74	36.00	50.25	97.33
	GCN-Transformer	10.58	44.19	68.23	94.82
	AGCN-Transformer	1.40	19.59	24.79	98.76

表3

图5

图6

表4

对比实验预测指标

月份	方法	RSE/%	MAE /kW	RMSE/kW	$R C O R R$ /%
1	T-GCN	6.30	56.61	86.08	96.82
	STGCN	4.28	52.86	65.99	98.13
	GCN-LSTM	6.32	56.21	72.20	96.24
	GCN-GRU	5.71	50.29	68.67	97.83
	MTGNN	2.45	39.25	44.65	98.42
	AGCN-Transformer	1.72	30.64	37.69	99.36
4	T-GCN	9.32	53.59	71.55	95.86
	STGCN	5.18	41.50	53.36	96.93
	GCN-LSTM	7.96	52.32	66.13	97.26
	GCN-GRU	4.62	37.65	47.96	97.65
	MTGNN	3.19	30.35	41.89	97.96
	AGCN-Transformer	3.13	33.87	41.48	98.43
7	T-GCN	7.11	39.65	61.05	96.76
	STGCN	8.39	41.28	66.30	96.02
	GCN-LSTM	7.57	48.24	62.97	96.54
	GCN-GRU	5.76	32.09	49.93	97.15
	MTGNN	5.77	35.10	48.42	96.98
	AGCN-Transformer	5.29	40.99	52.66	97.48
10	T-GCN	17.20	62.16	87.01	92.04
	STGCN	11.86	51.34	72.27	94.63
	GCN-LSTM	5.99	42.73	51.36	97.62
	GCN-GRU	4.34	28.04	43.71	97.95
	MTGNN	7.14	35.34	56.06	96.43
	AGCN-Transformer	1.40	19.58	24.79	98.76

表4

图7

参考文献 17

[1]	娄奇鹤, 刘虎, 谢祥颖, 等. 融合级联LSTM的短期电力负荷预测[J]. 计算机工程与应用, 2021, 57(18): 275-280.
	Lou Qihe, Liu Hu, Xie Xiangying, et al. Short-term Power Load Forecasting Using Cascaded LSTM Models[J]. Computer Engineering and Applications, 2021, 57(18): 275-280.
[2]	苏子越, 柴琳, 谢亮, 等. 基于变分模态分解的综合能源系统短期电负荷预测[J]. 热力发电, 2024, 53(12): 21-28.
	Su Ziyue, Chai Lin, Xie Liang, et al. Short-term Electrical Load Forecasting for Integrated Energy System Based on Variational Mode Decomposition[J]. Thermal Power Generation, 2024, 53(12): 21-28.
[3]	杨国华, 郑豪丰, 张鸿皓, 等. 基于Holt-Winters指数平滑和时间卷积网络的短期负荷预测[J]. 电力系统自动化, 2022, 46(6): 73-82.
	Yang Guohua, Zheng Haofeng, Zhang Honghao, et al. Short-term Load Forecasting Based on Holt-winters Exponential Smoothing and Temporal Convolutional Network[J]. Automation of Electric Power Systems, 2022, 46(6): 73-82.
[4]	杨凌, 彭文英, 杨思怡, 等. 基于极限学习机的短期电力负荷在线预测[J]. 兰州大学学报(自然科学版), 2024, 60(5): 637-644.
	Yang Ling, Peng Wenying, Yang Siyi, et al. Short-term Electricity Load Online Forecasting Based on Extreme Learning Machine[J]. Journal of Lanzhou University(Natural Sciences), 2024, 60(5): 637-644.
[5]	陈逸枞, 张大海, 李宇欣, 等. 基于DWT-MOSMA-SVM的多目标优化短期母线负荷预测[J]. 电力建设, 2023, 44(3): 49-55.
	Chen Yicong, Zhang Dahai, Li Yuxin, et al. Multi-objective Optimization Based on DWT-MOSMA-SVM for Short-term Bus Load Forecasting[J]. Electric Power Construction, 2023, 44(3): 49-55.
[6]	鲍海波, 吴阳晨, 张国应, 等. 基于特征加权Stacking集成学习的净负荷预测方法[J]. 电力建设, 2022, 43(9): 104-116.
	Bao Haibo, Wu Yangchen, Zhang Guoying, et al. Net Load Forecasting Method Based on Feature-weighted Stacking Ensemble Learning[J]. Electric Power Construction, 2022, 43(9): 104-116.
[7]	Tang Yihao, Cai Huafeng. Short-term Power Load Forecasting Based on VMD-pyraformer-adan[J]. IEEE Access, 2023, 11: 61958-61967.
[8]	俞胜, 孙可, 蔡华, 等. 结合极端梯度提升决策树与改进Informer的短期电力负荷预测方法[J]. 中国电力, 2025, 58(10): 195-205.
	Yu Sheng, Sun Ke, Cai Hua, et al. A Short-term Power Load Forecasting Method Combining Extreme Gradient Boosting Decision Tree with an Improved Informer[J]. Electric Power, 2025, 58(10): 195-205.
[9]	Rubasinghe Osaka, Zhang Tingze, Zhang Xinan, et al. Highly Accurate Peak and Valley Prediction Short-term Net Load Forecasting Approach Based on Decomposition for Power Systems with High PV Penetration[J]. Applied Energy, 2023, 333: 120641.
[10]	欧阳静, 杨吕, 尹康, 等. 基于ALIF-LSTM多任务学习的综合能源系统短期负荷预测[J]. 太阳能学报, 2022, 43(9): 499-507.
	Ouyang Jing, Yang Lü, Yin Kang, et al. Short-term Load Forecasting Method for Integrated Energy System Based on ALIF-LSTM and Multi-task Learning[J]. Acta Energiae Solaris Sinica, 2022, 43(9): 499-507.
[11]	He Yan, Zhang Hongli, Dong Yingchao, et al. Residential Net Load Interval Prediction Based on Stacking Ensemble Learning[J]. Energy, 2024, 296: 131134.
[12]	赵洪山, 吴雨晨, 潘思潮, 等. 基于TCN和高斯过程残差建模学习的净负荷概率预测方法[J]. 太阳能学报, 2024, 45(12): 588-595.
	Zhao Hongshan, Wu Yuchen, Pan Sichao, et al. Probabilistic Net Load Forecasting Based on TCN and Gaussian Process-enabled Residual Modeling Learning Approach[J]. Acta Energiae Solaris Sinica, 2024, 45(12): 588-595.
[13]	李丽芬, 张近月, 曹旺斌, 等. 基于双重注意力时间卷积长短期记忆网络的短期负荷预测[J]. 系统仿真学报, 2025, 37(8): 2004-2015.
	Li Lifen, Zhang Jinyue, Cao Wangbin, et al. Short-term Load Forecasting Based on Dual-attention Temporal Convolutional Long Short-term Memory Network[J]. Journal of System Simulation, 2025, 37(8): 2004-2015.
[14]	Kim Tae-Geun, Yoon Sung-Guk, Song Kyung-Bin. Very Short-term Load Forecasting Model for Large Power System Using GRU-attention Algorithm[J]. Energies, 2025, 18(13): 3229.
[15]	李芬, 李雨欣, 王亚维, 等. 考虑屋顶光伏热效应的短期净负荷预测[J]. 电力系统自动化, 2025, 49(9): 146-156.
	Li Fen, Li Yuxin, Wang Yawei, et al. Short-term Net Load Forecasting Considering Thermal Effect of Rooftop Photovoltaic[J]. Automation of Electric Power Systems, 2025, 49(9): 146-156.
[16]	董雷, 陈振平, 韩富佳, 等. 基于图卷积神经网络与K-means聚类的居民用户集群短期负荷预测[J]. 电网技术, 2023, 47(10): 4291-4301.
	Dong Lei, Chen Zhenping, Han Fujia, et al. Short-term Load Forecasting of Residential User Groups Based on Graph Convolutional Neural Network and K-means Clustering[J]. Power System Technology, 2023, 47(10): 4291-4301.
[17]	Vaswani A, Shazeer N, Parmar N, et al. Attention Is All You Need[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2017: 6000-6010.

月分	RSE/%		MAE/kW		RMSE/kW
月分	Q	X	Q	X	Q	X
1	2.35	1.72	34.53	30.64	44.05	37.69
4	5.84	3.13	47.91	33.88	56.65	41.49
7	7.39	5.29	49.58	41.00	62.24	52.66
10	2.39	1.41	26.10	19.59	32.44	24.79