Short-term Load Forecasting Based on Dual-attention Temporal Convolutional Long Short-term Memory Network

doi:10.16182/j.issn1004731x.joss.24-0282

Abstract

Abstract:

In order to improve the accuracy of load forecasting and fully extract the hidden relationships between load and other characteristic factors, a load forecasting method based on dual-attention temporal convolutional LSTM network (DA-TCLSNet) was proposed. Correlation analysis was conducted on the dataset using the maximum information coefficient method to perform feature screening to reduce the computational cost of the model. The model input was constructed using a sliding window. The DA-TCLSNet forecastingmodel was constructed. The temporal convolutional layer extracted dependencies at different time scales and captured the nonlinear characteristics among variables such as load and weather. The multi-head sparse self-attention layer focused on important information. The LSTM network layer explored the long-term dependency of time series. The temporal pattern attention layer realized adaptive learning of the relationships among different variables at the same time step and connected the above modules through a residual structure to improve the expressive power of the model. Experimental results show that the proposed algorithm offers better forecasting performance than other load forecasting methods.

Key words: load forecasting, temporal convolutional network, attention, residual structure, correlation analysis

CLC Number:

TP391.9

Li Lifen, Zhang Jinyue, Cao Wangbin, Mei Huawei. 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.

Figures/Tables 20

Fig. 1

Fig. 2

Fig. 3

Fig. 4

Fig. 5

Fig. 6

Fig. 7

Fig. 8

Fig. 9

Fig. 10

Table 1

Fig. 11

Table 2

Fig. 12

Fig. 13

Fig. 14

Table 3

Table 4

Table 5

Table 6

References 16

[1]	韩富佳, 王晓辉, 乔骥, 等. 基于人工智能技术的新型电力系统负荷预测研究综述[J]. 中国电机工程学报, 2023, 43(22): 8569-8591.
	Han Fujia, Wang Xiaohui, Qiao Ji, et al. Review on Artificial Intelligence Based Load Forecasting Research for the New-type Power System[J]. Proceedings of the CSEE, 2023, 43(22): 8569-8591.
[2]	刘雨佳, 樊艳芳, 白雪岩, 等. 基于特征交叉机制和误差补偿的风力发电功率短期预测[J]. 电工技术学报, 2023, 38(12): 3277-3288.
	Liu Yujia, Fan Yanfang, Bai Xueyan, et al. Short-term Wind Power Prediction Based on Feature Crossover Mechanism and Error Compensation[J]. Transactions of China Electrotechnical Society, 2023, 38(12): 3277-3288.
[3]	乔石, 王磊, 张鹏超, 等. 基于模态分解及注意力机制长短时间网络的短期负荷预测[J]. 电网技术, 2022, 46(10): 3940-3951.
	Qiao Shi, Wang Lei, Zhang Pengchao, et al. Short-term Load Forecasting by Long-and Short-term Temporal Networks with Attention Based on Modal Decomposition[J]. Power System Technology, 2022, 46(10): 3940-3951.
[4]	Jiang Huaiguang, Zhang Yingchen, Muljadi E, et al. A Short-term and High-resolution Distribution System Load Forecasting Approach Using Support Vector Regression with Hybrid Parameters Optimization[J]. IEEE Transactions on Smart Grid, 2018, 9(4): 3341-3350.
[5]	孙超, 吕奇, 朱思曈, 等. 基于双层XGBoost算法考虑多特征影响的超短期电力负荷预测[J]. 高电压技术, 2021, 47(8): 2885-2895.
	Sun Chao, Qi Lü, Zhu Sitong, et al. Ultra-short-term Power Load Forecasting Based on Two-layer XGBoost Algorithm Considering the Influence of Multiple Features[J]. High Voltage Engineering, 2021, 47(8): 2885-2895.
[6]	臧海祥, 许瑞琦, 刘璟璇, 等. 基于多维融合特征和卷积神经网络的多任务用户短期负荷预测[J]. 电力系统自动化, 2023, 47(13): 69-77.
	Zang Haixiang, Xu Ruiqi, Liu Jingxuan, et al. Short-term Load Forecasting for Multi-task Consumers Based on Multi-dimensional Fusion Feature and Convolutional Neural Network[J]. Automation of Electric Power Systems, 2023, 47(13): 69-77.
[7]	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.
[8]	Guo Yixiu, Li Yong, Qiao Xuebo, et al. BiLSTM Multitask Learning-based Combined Load Forecasting Considering the Loads Coupling Relationship for Multienergy System[J]. IEEE Transactions on Smart Grid, 2022, 13(5): 3481-3492.
[9]	Jayashankara M, Shah Priyansh, Sharma Anshul, et al. A Novel Approach for Short-term Energy Forecasting in Smart Buildings[J]. IEEE Sensors Journal, 2023, 23(5): 5307-5314.
[10]	刘国海, 孙文卿, 吴振飞, 等. 基于Attention-GRU的短期光伏发电功率预测[J]. 太阳能学报, 2022, 43(2): 226-232.
	Liu Guohai, Sun Wenqing, Wu Zhenfei, et al. Short-term Photovoltaic Power Forecasting Based on Attention-GRU Model[J]. Acta Energiae Solaris Sinica, 2022, 43(2): 226-232.
[11]	Imani Maryam. Electrical Load-temperature CNN for Residential Load Forecasting[J]. Energy, 2021, 227: 120480.
[12]	王琛, 王颖, 郑涛, 等. 基于ResNet-LSTM网络和注意力机制的综合能源系统多元负荷预测[J]. 电工技术学报, 2022, 37(7): 1789-1799.
	Wang Chen, Wang Ying, Zheng Tao, et al. Multi-energy Load Forecasting in Integrated Energy System Based on ResNet-LSTM Network and Attention Mechanism[J]. Transactions of China Electrotechnical Society, 2022, 37(7): 1789-1799.
[13]	王炜, 冯斌, 黄刚, 等. 基于自注意力编码器和深度神经网络的短期净负荷预测[J]. 中国电机工程学报, 2023, 43(23): 9072-9083.
	Wang Wei, Feng Bin, Huang Gang, et al. Short-term Net Load Forecasting Based on Self-attention Encoder and Deep Neural Network[J]. Proceedings of the CSEE, 2023, 43(23): 9072-9083.
[14]	Dong Ming, Grumbach L. A Hybrid Distribution Feeder Long-term Load Forecasting Method Based on Sequence Prediction[J]. IEEE Transactions on Smart Grid, 2020, 11(1): 470-482.
[15]	Reshef D N, Reshef Y A, Finucane H K, et al. Detecting Novel Associations in Large Data sets[J]. Science, 2011, 334(6062): 1518-1524.
[16]	Shih S Y, Sun Fankeng, Lee H Y. Temporal Pattern Attention for Multivariate Time Series Forecasting[J]. Machine Learning, 2019, 108(8): 1421-1441.

模块	参数类型	参数值
TCN	膨胀因子	[1, 2]
	膨胀层	2
	隐藏层单元数	[64, 16]
多头稀疏自注意力	注意头	8
	注意力因子	5
	隐藏单元数	96
时间模式注意力	隐藏单元数	10
LSTM	隐藏单元数	100

数据集	模型	MAPE/%	RMSE/MW	MAE/MW	R²
NH	LSTM	1.60	28.74	21.31	0.991 2
	GRU	1.62	30.58	22.41	0.990 0
	XGBoost	1.61	30.37	22.26	0.990 2
	SVR	1.81	33.02	25.00	0.988 4
	DA-TCLSNet	0.91	17.96	12.70	0.996 5
SEMASS	LSTM	1.33	32.98	24.52	0.993 8
	GRU	1.32	33.46	24.24	0.993 6
	XGBoost	1.82	43.72	32.29	0.989 1
	SVR	1.52	34.76	27.54	0.993 1
	DA-TCLSNet	0.99	26.58	18.32	0.996 0
Panama	LSTM	2.31	32.88	23.86	0.963 5
	GRU	2.31	32.33	23.94	0.964 7
	XGBoost	2.26	31.35	23.83	0.966 8
	SVR	2.42	32.70	24.67	0.963 9
	DA-TCLSNet	1.81	26.00	18.92	0.977 2

数据类型	模型	MAPE/%	RMSE/MW	MAE/MW	R²
平稳序列	LSTM	1.57	26.57	19.29	0.987 1
	GRU	1.70	29.73	21.37	0.983 8
	XGBoost	1.47	23.97	18.27	0.989 5
	SVR	1.42	22.79	17.79	0.990 5
	DA-TCLSNet	0.86	15.24	10.93	0.995 8
波动序列	LSTM	1.72	31.98	23.64	0.987 6
	GRU	1.59	30.87	22.24	0.988 4
	XGBoost	1.71	34.62	24.19	0.985 4
	SVR	2.21	40.73	30.57	0.979 8
	DA-TCLSNet	0.82	16.56	11.54	0.996 7

数据类型	模型	MAPE/%	RMSE/MW	MAE/MW	R²
平稳序列	LSTM	1.36	32.82	23.61	0.990 7
	GRU	1.31	32.39	22.83	0.990 9
	XGBoost	1.82	38.01	30.28	0.987 5
	SVR	1.54	32.34	27.09	0.990 9
	DA-TCLSNet	0.89	21.93	15.45	0.995 8
波动序列	LSTM	1.38	35.15	25.22	0.993 2
	GRU	1.43	35.70	25.74	0.993 0
	XGBoost	2.41	53.42	42.07	0.984 3
	SVR	1.70	38.73	29.62	0.991 8
	DA-TCLSNet	1.20	29.97	22.21	0.995 1

模型	MAPE/%	RMSE/MW	MAE/MW	R²
LSTM	2.37	34.91	24.07	0.957 7
GRU	2.44	35.34	24.81	0.956 6
XGBoost	2.47	36.32	26.35	0.954 2
SVR	3.22	41.49	32.55	0.940 2
DA-TCLSNet	1.96	28.77	20.24	0.971 2