结合贝叶斯Autoformer的多维自适应短期电力负荷概率预测方法

周师琦; 王俊帆; 赖俊升; 袁毓杰; 董哲康

doi:10.11999/JEIT240398

结合贝叶斯Autoformer的多维自适应短期电力负荷概率预测方法

doi: 10.11999/JEIT240398

周师琦^{1, 2},
王俊帆^{1, 2},
赖俊升³,
袁毓杰^4, ,,
董哲康^{1, 2}

1.
杭州电子科技大学电子信息学院杭州 310018
2.
浙江省装备电子重点实验室杭州 310018
3.
英国伦敦布鲁奈尔大学电子与计算机工程系伦敦 UB8 3PH
4.
中国民航大学空中交通管理学院天津 300300

基金项目: 国家自然科学基金(62206062)，长三角科技创新共同体联合攻关重点项目(2023CSJGG1300)，浙江省属高校基本科研业务费项目(GK229909299001-06)

详细信息

作者简介:
周师琦：女，博士生，研究方向为智能电网、能源管理、深度学习

王俊帆：女，博士生，研究方向为能源管理、深度学习

赖俊升：男，教授，研究方向为智能电网、能源管理

袁毓杰：女，讲师，研究方向为智能电网、能源管理

董哲康：男，副教授，研究方向为能源管理、深度学习

通讯作者:
袁毓杰　18114026@bjtu.edu.cn

中图分类号: TM715; TP18
计量
- 文章访问数: 154
- HTML全文浏览量: 72
- PDF下载量: 13
- 被引次数: 0
出版历程
- 收稿日期: 2024-05-21
- 修回日期: 2024-08-26
- 网络出版日期: 2024-08-30

Multi-view Adaptive Probabilistic Load Forecasting Combing Bayesian Autoformer Network

ZHOU Shiqi^{1, 2},
WANG Junfan^{1, 2},
LAI Junsheng³,
YUAN Yujie^{4
, ,},
DONG Zhekang^{1, 2}

1.
School of Electronics and Information, Hangzhou Dianzi University, Hangzhou 310018, China
2.
Zhejiang Province Key Lab of Equipment Electronics, Hangzhou 310018, China
3.
Department of Electronic and Computer Engineering, Brunel University London, London UB8 3PH, UK
4.
School of Air Traffic Management, Civil Aviation University of China, Tianjin 300300, China

Funds: The National Natural Science Foundation of China (62206062), Yangtze River Delta Science and Technology Innovation Community Jointly Tackled Key Project (2023CSJGG1300), The Fundamental Research Funds for the Provincial University of Zhejiang (GK229909299001-06)

摘要

摘要: 建立准确的电力负荷短期预测模型对于电力系统的稳定运行和智能化进程至关重要。目前的主流预测方法无法很好地突破数据波动性和模型不确定性两个问题。基于此，该文提出一种基于贝叶斯Autoformer的多维自适应短期电力负荷概率预测方法。具体地，提出自适应特征提取方法获取多维度特征，通过捕捉多尺度特征和时频局部信息，增强模型对负荷数据中高波动性和非线性特征的处理能力。其次，提出基于贝叶斯Autoformer的预测模型，它可以捕获负荷数据中重要子序列特征以及不确定性，并通过贝叶斯优化方法实现概率预测分布和参数分布的动态更新。所提模型在3个量级(GW, MW, KW)的实际负荷数据集上进行一系列实验分析(对比分析、自适应分析、鲁棒性分析)。结果表明，所提预测模型在自适应和准确性方面具有优越的性能，均方根误差(RMSE)、弹球损失(Pinball Loss)、连续概率评分(CRPS)，相较对比方法分别提升1.9%, 24.2%, 4.5%。
- 负荷预测 /
- 概率预测 /
- 贝叶斯神经网络 /
- Autoformer
Abstract: Establishing accurate short-term forecasting models for electrical load is crucial for the stable operation and intelligent advancement of power systems. Traditional methods have not adequately addressed the issues of data volatility and model uncertainty. In this paper, a multi-dimensional adaptive short-term forecasting method for electrical load based on Bayesian Autoformer network is proposed. Specifically, an adaptive feature selection method is designed to capture multi-dimensional features. By capturing multi-scale features and time-frequency localized information, the model is enhanced to handle high volatility and nonlinear features in load data. Subsequently, an adaptive probabilistic forecasting model based on Bayesian Autoformer network is proposed. It captures relationships of significant subsequence features and associated uncertainties in load time series data, and dynamically updates the probability prediction model and parameter distributions through Bayesian optimization. The proposed model is subjected to a series of experimental analyses (comparative analysis, adaptive analysis, robustness analysis) on real load datasets of three different magnitudes (GW, MW, and KW). The model exhibits superior performance in adaptability and accuracy, with average improvements in Root Mean Square Error (RMSE), Pinball Loss, and Continuous Ranked Probability Score (CRPS) of 1.9%, 24.2%, and 4.5%, respectively.
- Load forecasting /
- Probabilistic forecasting /
- Bayesian neural network /
- Autoformer

HTML全文

图 1 电力负荷数据的波动性分析

下载: 全尺寸图片幻灯片

图 2 电力负荷数据在不同时期的分布情况

下载: 全尺寸图片幻灯片

图 3 电力负荷数据的漂移检测结构

下载: 全尺寸图片幻灯片

图 4 基于贝叶斯 Autoformer 的多维自适应电负荷概率预测模型

下载: 全尺寸图片幻灯片

图 5 超参数设置

下载: 全尺寸图片幻灯片

图 6 3 个电力负荷数据集在不同预测维度下概率预测的结果

下载: 全尺寸图片幻灯片

图 7 不同特征选择方法下的预测结果

下载: 全尺寸图片幻灯片

表 1 模型的超参数设置

参数	值
编码层数l_E	4
解码层数l_D	4
多头注意力h	8
模型维度d_M	24
滑动窗口长度l_W	168
滑动窗口步长d_W	1

下载: 导出CSV

表 2 不同预测模型在3 个数据集下的性能对比结果

方法	New England (GW)			ASU (MW)			400builds (KW)
方法	RMSE	Pinball	CRPS	RMSE	Pinball	CRPS	RMSE	Pinball	CRPS
FPSwq2Q^[10]	1.065 9	0.194 7	0.998 1	0.176 7	0.014 7³	0.207 1	0.069 1	0.019 6	0.059 2
QLSTM^[8]	0.985 7³	0.117 9²	0.692 4	0.179 3	0.016 2	0.200 7	0.062 3	0.013 7	0.061 9
*BNN^[15]	1.217 7	0.161 8	0.595 7³	0.182 2	0.016 7	0.195 9	0.140 8	0.015 9	0.094 2
*BSDeT^[14]	1.032 4	0.148 2	0.974 3	0.148 9³	0.013 7²	0.130 8²	0.058 2³	0.010 9³	0.045 2²
**MQR^[12]	1.527 3	0.216 9	0.631 0	0.174 9	0.016 8	0.201 5	0.091 6	0.020 7	0.072 8
**APLF^[13]	0.911 0²	0.132 8³	0.506 7²	0.140 6²	0.018 2	0.144 7³	0.032 2²	0.010 5²	0.048 1³
本文	0.910 9¹	0.078 0¹	0.495 1¹	0.139 4¹	0.012 4¹	0.120 1¹	0.031 4¹	0.010 3¹	0.023 7¹
注：上标1,2,3分别表示排名第1、第2和第3；基于BNN的概率预测模型，*为自适应概率预测模型

下载: 导出CSV

表 3 不同注意力机制用于贝叶斯概率预测的比较

方法	New England			ASU			400 builds
方法	RMSE	Pinball	CRPS	RMSE	Pinball	CRPS	RMSE	Pinball	CRPS
BNN+Full	0.943 5	0.082 5	0.790 5	0.190 8	0.015 1	0.147 6	0.032 5	0.010 9	0.024 9
BNN+Log	0.930 2	0.081 0	0.778 5	0.189 8	0.015 1	0.146 6	0.0322	0.0107	0.0244
BNN+In	0.943 5	0.079 5	0.763 5	0.190 3	0.014 9¹	0.146 2	0.031 7	0.011 2	0.024 1
BNN+Auto	0.915 0¹	0.078 0¹	0.742 5¹	0.188 9¹	0.014 9¹	0.144 2¹	0.031 4¹	0.010 3¹	0.023 9¹
注：下标1表示排名第1。

下载: 导出CSV

表 4 概率预测模型的稳定性分析

评价指标		RMSE	Pinball	CRPS
New England	对照组	0.9150	0.0780	0.7425
New England	噪声组	0.9157	0.0784	0.7431
ASU	对照组	0.1889	0.0149	0.1442
ASU	噪声组	0.1890	0.0150	0.1446
400build	对照组	0.0314	0.0103	0.0239
400build	噪声组	0.0318	0.0107	0.0240

下载: 导出CSV

参考文献(24)

[1]	LAI C S, YANG Yuxiang, PAN Keda, et al. Multi-view neural network ensemble for short and mid-term load forecasting[J]. IEEE Transactions on Power Systems, 2021, 36(4): 2992–3003. doi: 10.1109/TPWRS.2020.3042389.
[2]	KIM N, PARK H, LEE J, et al. Short-term electrical load forecasting with multidimensional feature extraction[J]. IEEE Transactions on Smart Grid, 2022, 13(4): 2999–3013. doi: 10.1109/TSG.2022.3158387.
[3]	LI Jian, DENG Daiyu, ZHAO Junbo, 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. doi: 10.1109/TII.2020.3000184.
[4]	DANG Sanlei, PENG Long, ZHAO Jingming, et al. A quantile regression random forest-based short-term load probabilistic forecasting method[J]. Energies, 2022, 15(2): 663. doi: 10.3390/en15020663.
[5]	王俊帆, 陈毅, 高明煜, 等. 智能交通感知新范式: 面向元宇宙的交通标志检测架构[J]. 电子与信息学报, 2024, 46(3): 777–789. doi: 10.11999/JEIT230357. WANG Junfan, CHEN Yi, GAO Mingyu, et al. A new paradigm for intelligent traffic perception: A traffic sign detection architecture for the metaverse[J]. Journal of Electronics and Information Technology, 2024, 46(3): 777–789. doi: 10.11999/JEIT230357.
[6]	WANG Bingzhi, MAZHARI M, and CHUNG C Y. A novel hybrid method for short-term probabilistic load forecasting in distribution networks[J]. IEEE Transactions on Smart Grid, 2022, 13(5): 3650–3661. doi: 10.1109/TSG.2022.3171499.
[7]	ZHANG Dongxue, WANG Shuai, LIANG Yuqiu, et al. A novel combined model for probabilistic load forecasting based on deep learning and improved optimizer[J]. Energy, 2023, 264: 126172. doi: 10.1016/j.energy.2022.126172.
[8]	WANG Yi, GAN Dahua, SUN Mingyang, et al. Probabilistic individual load forecasting using pinball loss guided LSTM[J]. Applied Energy, 2019, 235: 10–20. doi: 10.1016/j.apenergy.2018.10.078.
[9]	LUO Long, DONG Jizhe, KONG Weizhe, et al. Short-term probabilistic load forecasting using quantile regression neural network with accumulated hidden layer connection structure[J]. IEEE Transactions on Industrial Informatics, 2024, 20(4): 5818–5828. doi: 10.1109/TII.2023.3341242.
[10]	FAUSTINE A and PEREIRA L. FPSeq2Q: Fully parameterized sequence to quantile regression for net-load forecasting with uncertainty estimates[J]. IEEE Transactions on Smart Grid, 2022, 13(3): 2440–2451. doi: 10.1109/TSG.2022.3148699.
[11]	LEMOS-VINASCO J, BACHER P, and MØLLER J K. Probabilistic load forecasting considering temporal correlation: Online models for the prediction of households’ electrical load[J]. Applied Energy, 2021, 303: 117594. doi: 10.1016/j.apenergy.2021.117594.
[12]	BRACALE A, CARAMIA P, DE FALCO P, et al. Multivariate quantile regression for short-term probabilistic load forecasting[J]. IEEE Transactions on Power Systems, 2020, 35(1): 628–638. doi: 10.1109/TPWRS.2019.2924224.
[13]	ÁLVAREZ V, MAZUELAS S, and LOZANO J A. Probabilistic load forecasting based on adaptive online learning[J]. IEEE Transactions on Power Systems, 2021, 36(4): 3668–3680. doi: 10.1109/TPWRS.2021.3050837.
[14]	YU Binbin, LI Jianjing, LIU Che, et al. A novel short-term electrical load forecasting framework with intelligent feature engineering[J]. Applied Energy, 2022, 327: 120089. doi: 10.1016/j.apenergy.2022.120089.
[15]	BRUSAFERRI A, MATTEUCCI M, SPINELLI S, et al. Probabilistic electric load forecasting through Bayesian mixture density networks[J]. Applied Energy, 2022, 309: 118341. doi: 10.1016/j.apenergy.2021.118341.
[16]	ISO NEW England[EB/OL]. https://www.iso-ne.com, 2024.
[17]	Metabolism[OL]. https://cm.asu.edu, 2023.
[18]	ANAO[OL]. https://www.anao.gov.au/work/performance-audit, 2023.
[19]	FEKRI M N, PATEL H, GROLINGER K, et al. Deep learning for load forecasting with smart meter data: Online adaptive recurrent neural network[J]. Applied Energy, 2021, 282: 116177. doi: 10.1016/j.apenergy.2020.116177.
[20]	FAN Guofeng, PENG Liling, and HONG W C. Short-term load forecasting based on empirical wavelet transform and random forest[J]. Electrical Engineering, 2022, 104(6): 4433–4449. doi: 10.1007/s00202-022-01628-y.
[21]	董哲康, 钱智凯, 周广东, 等. 基于忆阻的全功能巴甫洛夫联想记忆电路的设计、实现与分析[J]. 电子与信息学报, 2022, 44(6): 2080–2092. doi: 10.11999/JEIT210376. DONG Zhekang, QIAN Zhikai, ZHOU Guangdong, et al. Memory circuit design, implementation and analysis based on Memristor full-function Pavlov associative[J]. Journal of Electronics and Information Technology, 2022, 44(6): 2080–2092. doi: 10.11999/JEIT210376.
[22]	CHEN Minghao, PENG Houwen, FU Jianlong, et al. AutoFormer: Searching transformers for visual recognition[C]. 2021 IEEE/CVF International Conference on Computer Vision, Montreal, Canada, 2021: 12270–12280. doi: 10.1109/ICCV48922.2021.01205.
[23]	CAO Defu, WANG Yujing, DUAN Juanyong, et al. Spectral temporal graph neural network for multivariate time-series forecasting[C]. Proceedings of the 34th International Conference on Neural Information Processing Systems, Vancouver, Canada, 2020: 1491.
[24]	LI Lei, XUE Zhigang, and DU Xiuquan. ASCRB: Multi-view based attentional feature selection for CircRNA-binding site prediction[J]. Computers in Biology and Medicine, 2023, 162: 107077. doi: 10.1016/j.compbiomed.2023.107077.