基于迁移学习和TCN-BiGRU的短期负荷预测

刘杰; 周博文; 田明; 韩轲

doi:10.13700/j.bh.1001-5965.2024.0056

基于迁移学习和TCN-BiGRU的短期负荷预测

doi: 10.13700/j.bh.1001-5965.2024.0056

刘杰¹,
周博文¹,
田明²,
韩轲^3, ,

1.
哈尔滨理工大学黑龙江省模式识别与信息感知重点实验室，哈尔滨 150080
2.
中国电信股份有限公司黑龙江分公司，哈尔滨 150040
3.
哈尔滨商业大学计算机与信息工程学院，哈尔滨 150028

基金项目:

黑龙江省自然科学基金(LH2020F009)

详细信息

通讯作者:
E-mail：thruster@163.com

中图分类号: V247；TP391
计量
- 文章访问数: 403
- HTML全文浏览量: 121
- PDF下载量: 34
- 被引次数: 0
出版历程
- 收稿日期: 2024-01-23
- 录用日期: 2024-03-26
- 网络出版日期: 2024-04-07
- 整期出版日期: 2026-04-30

Short-term load forecasting based on transfer learning and TCN-BiGRU

LIU Jie¹,
ZHOU Bowen¹,
TIAN Ming²,
HAN Ke^{3
, ,}

1.
Heilongjiang Province Key Laboratory of Pattern Recognition and Information Perception，Harbin University of Science and Technology，Harbin 150080，China
2.
China Telecom Heilongjiang Branch，Harbin 150040，China
3.
School of Computer and Information Engineering，Harbin University of Commerce，Harbin 150028，China

Funds:

Natural Science Foundation of Heilongjiang Province (LH2020F009)

More Information

Corresponding author: E-mail：thruster@163.com

摘要

摘要:
电力负荷预测对于电力系统的稳定运行有着重要意义。传统短期预测算法常使用线性回归模型进行负荷预测，无法捕捉到复杂的负荷波动，导致预测结果准确性受到限制，因此，提出一种基于迁移学习(TL)的时间卷积网络-双向门控循环单元(TCN-BiGRU)模型。采用迁移学习策略将相关性高的信息迁移到实验模型中，利用K-medoids聚类算法对数据进行聚类分析，通过并行卷积策略提取TCN不同尺度的特征，利用时间注意力(TA)捕获相关信息，结合BiGRU进一步提取TCN训练输出的非线性特征，使用动态多群粒子群优化(DMS-PSO)算法对网络训练的超参数寻找最佳的超参数组合。实验结果表明：相对于门控循环单元(GRU)，所提TL-TCN-BiGRU算法的平均绝对误差(MAE)降低了38.6%，均方根误差(RMSE)降低了40.7%，平均绝对百分比误差(MAPE)降低了30.4%，R²提升了5.3%。
- 短期负荷预测 /
- 迁移学习 /
- 时间卷积网络 /
- K-medoids聚类 /
- 融合网络
Abstract:
Electricity load forecasting is of great significance to the stable operation of power systems. For load forecasting, traditional short-term forecasting techniques frequently employ linear regression models, which have low forecasting accuracy due to the models’ inability to incorporate complicated load changes. A temporal convolutional network-bidirectional gated recurrent unit (TCN-BiGRU) model based on transfer learning (TL) is proposed. Highly relevant information is moved into the experimental model using a transfer learning strategy; the data is clustered and analyzed using a K-medoids clustering algorithm; features at various TCN scales are extracted using a parallel convolution strategy; pertinent information is captured using temporal attention (TA); and the TCN training is further extracted using a BiGRU. The non-linear features of the output are further extracted using the dynamic multigroup particle swarm optimization (DMS-PSO) algorithm to optimize and tune the hyperparameters of the network training in order to find the best combination of hyperparameters. The experimental results show that the proposed TL-TCN-BiGRU algorithm reduces mean absolute error (MAE) by 38.6%, root mean square error (RMSE) by 40.7%, mean absolute percentage error (MAPE) by 30.4%, and R² by 5.3% relative to the gated recurrent unit (GRU).
- short-term load forecasting /
- transfer learning /
- temporal convolutional network /
- K-medoids clustering /
- fusion network

HTML全文

图 1 TA-TCN结构设计

Figure 1. TA-TCN structural design

下载: 全尺寸图片幻灯片

图 2 TL-TCN-BiGRU模型整体框架

Figure 2. Overall framework of TL-TCN-BiGRU model

下载: 全尺寸图片幻灯片

图 3 特征相关性热力图

Figure 3. Characteristic correlation heat map

下载: 全尺寸图片幻灯片

图 4 不同簇数量下SSE和轮廓系数的变化曲线

Figure 4. Curves of SSE and silhouette coefficient under different cluster numbers

下载: 全尺寸图片幻灯片

图 5 聚类数为4时的负荷曲线

Figure 5. Load profile for a clustering number of 4

下载: 全尺寸图片幻灯片

图 6 不同模型评价结果

Figure 6. Different model evaluation result graphs

下载: 全尺寸图片幻灯片

图 7 不同模型24 h预测结果

Figure 7. 24 h prediction results of different models

下载: 全尺寸图片幻灯片

表 1 DMS-PSO算法寻优结果

Table 1. DMS-PSO algorithm optimisation results

寻优参数	范围	寻优结果
BiGRU隐藏层神经元数量	[1,50]	31
卷积核大小	[3,5]	4
Dropout	[0,0.5]	0.01
学习率	[0.0001,0.01]	0.001
批处理大小	[16,32]	20
迭代次数	[100,200]	121

下载: 导出CSV

表 2 不同聚类算法指标对比

Table 2. Comparison of indicators of different clustering methods

聚类算法	DB指数	CH指数
DBSCAN	1.459	805.715
K-means	1.368	990.687
K-medoids	1.231	1083.451

下载: 导出CSV

表 3 不同优化算法效果对比

Table 3. Comparison of effect of different optimization algorithms

聚类算法	MAE/kW	MAPE/‰
GA	0.092	0.7023
PSO	0.075	0.5236
DMS-PSO	0.035	0.4183

下载: 导出CSV

表 4 消融实验结果

Table 4. Ablation experiment results

TCN	TA结构	BiGRU	DMS-PSO	K-medoids聚类	TL模块	MAE/kW	RMSE/kW	MAPE/‰	$ {R}^{2} $
√						0.065	0.099	0.7435	0.908
	√					0.062	0.091	0.7174	0.922
	√	√				0.059	0.088	0.6709	0.928
	√	√	√			0.052	0.080	0.5795	0.940
	√	√	√	√		0.046	0.072	0.5061	0.964
√	√	√	√	√	√	0.035	0.054	0.4183	0.972

下载: 导出CSV

表 5 不同模型评价结果

Table 5. Evaluation results of different models

模型	MAE/kW	RMSE/kW	MAPE/‰	$ {R}^{2} $
BP	0.066	0.096	0.7995	0.914
SVR	0.076	0.101	1.1669	0.905
LSTM	0.058	0.097	0.6244	0.912
GRU	0.057	0.091	0.6009	0.923
TCN	0.065	0.099	0.7435	0.908
本文模型	0.035	0.054	0.4183	0.972

下载: 导出CSV

参考文献(25)

[1]	刘倩倩, 刘钰山, 温烨婷, 等. 基于PCC-LSTM模型的短期负荷预测方法[J]. 北京航空航天大学学报, 2022, 48(12): 2529-2536. LIU Q Q, LIU Y S, WEN Y T, et al. Short-term load forecasting method based on PCC-LSTM model[J]. Journal of Beijing University of Aeronautics and Astronautics, 2022, 48(12): 2529-2536(in Chinese).
[2]	LI X M, GUO X C, LIU L N, et al. A novel seasonal grey model for forecasting the quarterly natural gas production in China[J]. Energy Reports, 2022, 8: 9142-9157.
[3]	王艳松, 申晓阳, 李强, 等. 基于PCA-GRD-LWR模型的海上油田中长期最大电力负荷预测[J]. 中国石油大学学报(自然科学版), 2023, 47(2): 129-135. WANG Y S, SHEN X Y, LI Q, et al. Forecasting of medium and long-term maximum power load for offshore oilfields based on PCA-GRD-LWR model[J]. Journal of China University of Petroleum (Edition of Natural Science), 2023, 47(2): 129-135(in Chinese).
[4]	李丹, 孙光帆, 缪书唯, 等. 基于多维时序信息融合的短期电力负荷预测方法[J]. 中国电机工程学报, 2023, 43(S1): 94-106. LI D, SUN G F, MIAO S W, et al. A short-term power load forecasting method based on multidimensional temporal information fusion[J]. Proceedings of the CSEE, 2023, 43(S1): 94-106(in Chinese).
[5]	ZHANG R X, ZHU Z Y, YUAN M, et al. Regional residential short-term load-interval forecasting based on SSA-LSTM and load consumption consistency analysis[J]. Energies, 2023, 16(24): 8062.
[6]	BACANIN N, STOEAN C, ZIVKOVIC M, et al. On the benefits of using metaheuristics in the hyperparameter tuning of deep learning models for energy load forecasting[J]. Energies, 2023, 16(3): 1434.
[7]	AL-JAMIMI H A, BINMAKHASHEN G M, WORKU M Y, et al. Advancements in household load forecasting: deep learning model with hyperparameter optimization[J]. Electronics, 2023, 12(24): 4909.
[8]	WANG J Z, WANG K, LI Z W, et al. A multitask integrated deep-learning probabilistic prediction for load forecasting[J]. IEEE Transactions on Power Systems, 2024, 39(1): 1240-1250.
[9]	TRAN T N. Grid search of convolutional neural network model in the case of load forecasting[J]. Archives of Electrical Engineering, 2021, 70(1): 25-36.
[10]	ESKANDARI H, IMANI M, MOGHADDAM M P. Convolutional and recurrent neural network based model for short-term load forecasting[J]. Electric Power Systems Research, 2021, 195: 107173.
[11]	CAICEDO-VIVAS J S, ALFONSO-MORALES W. Short-term load forecasting using an LSTM neural network for a grid operator[J]. Energies, 2023, 16(23): 7878.
[12]	李艳波, 尹镨, 陈俊硕, 等. 结合改进残差网络和Bi-LSTM的短期电力负荷预测[J]. 哈尔滨工业大学学报, 2023, 55(8): 79-86. LI Y B, YIN P, CHEN J S, et al. Short-term power load forecasting based on combination of residual network and Bi-LSTM[J]. Journal of Harbin Institute of Technology, 2023, 55(8): 79-86(in Chinese).
[13]	BAI S J, KOLTER J Z, KOLTUN V. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling[EB/OL]. (2018-04-19)[2024-01-10]. https://arxiv.org/abs/1803.01271.
[14]	CHUNG J, ÇAGLAR G, CHO K, et al. Empirical evaluation of gated recurrent neural networks on sequence modeling[EB/OL]. (2014-12-11)[2024-01-10]. https://arxiv.org/abs/1412.3555.
[15]	XU H S, FAN G L, KUANG G F, et al. Construction and application of short-term and mid-term power system load forecasting model based on hybrid deep learning[J]. IEEE Access, 2023, 11: 37494-37507.
[16]	吴晨, 姚菁, 薛贵元, 等. 基于MMoE多任务学习和长短时记忆网络的综合能源系统负荷预测[J]. 电力自动化设备, 2022, 42(7): 33-39. WU C, YAO J, XUE G Y, et al. Load forecasting of integrated energy system based on MMoE multi-task learning and LSTM[J]. Electric Power Automation Equipment, 2022, 42(7): 33-39(in Chinese).
[17]	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 G H, ZHENG H F, ZHANG H H, 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(in Chinese).
[18]	叶林, 李奕霖, 裴铭, 等. 寒潮天气小样本条件下的短期风电功率组合预测[J]. 中国电机工程学报, 2023, 43(2): 543-555. YE L, LI Y L, PEI M, et al. Combined approach for short-term wind power forecasting under cold weather with small sample[J]. Proceedings of the CSEE, 2023, 43(2): 543-555(in Chinese).
[19]	邹智, 吴铁洲, 张晓星, 等. 基于贝叶斯优化CNN-BiGRU混合神经网络的短期负荷预测[J]. 高电压技术, 2022, 48(10): 3935-3945. ZOU Z, WU T Z, ZHANG X X, et al. Short-term load forecasting based on Bayesian optimized CNN-BiGRU hybrid neural network[J]. High Voltage Technology, 2022, 48(10): 3935-3945(in Chinese).
[20]	HUO F C, CHEN Y, REN W J, et al. Prediction of reservoir key parameters in ‘sweet spot’ on the basis of particle swarm optimization to TCN-LSTM network[J]. Journal of Petroleum Science and Engineering, 2022, 214: 110544.
[21]	苏伟, 肖小龙, 史明明, 等. 周期规律增强的多视角短期电力负荷预测[J]. 北京航空航天大学学报, 2024, 50(2): 477-486. SU W, XIAO X L, SHI M M, et al. Periodic pattern-enhanced multi-view short-term load prediction[J]. Journal of Beijing University of Aeronautics and Astronautics, 2024, 50(2): 477-486(in Chinese).
[22]	LI F, WANG C F. Develop a multi-linear-trend fuzzy information granule based short-term time series forecasting model with K-medoids clustering[J]. Information Sciences, 2023, 629: 358-375.
[23]	欧阳福莲, 王俊, 周杭霞. 基于改进迁移学习和多尺度CNN-BiLSTM-Attention的短期电力负荷预测方法[J]. 电力系统保护与控制, 2023, 51(2): 132-140. OUYANG F L, WANG J, ZHOU H X. Short-term power load forecasting method based on improved hierarchical transfer learning and multi-scale CNN-BiLSTM-Attention[J]. Power System Protection and Control, 2023, 51(2): 132-140(in Chinese).
[24]	HU Q H, ZHANG R J, ZHOU Y C. Transfer learning for short-term wind speed prediction with deep neural networks[J]. Renewable Energy, 2016, 85: 83-95.
[25]	JIN Y W, ACQUAH M A, SEO M, et al. Short-term electric load prediction using transfer learning with interval estimate adjustment[J]. Energy and Buildings, 2022, 258: 111846.