[1]沈佳眉,朱 红.基于注意力机制-集成学习耦合的电力时间序列数据补全方法[J].机械与电子,2024,42(11):3-10.
　SHEN Jiamei,ZHU Hong.Electric Power Time Series Data Imputation Method Based on Attention Mechanism and Ensemble Learning[J].Machinery & Electronics,2024,42(11):3-10.
点击复制

基于注意力机制-集成学习耦合的电力时间序列数据补全方法()
分享到：

《机械与电子》[ISSN:1001-2257/CN:52-1052/TH]

卷:
42
期数:
2024年11期
页码:
3-10
栏目:
研究与设计
出版日期:
2024-11-27

文章信息/Info

Title:
Electric Power Time Series Data Imputation Method Based on Attention Mechanism and Ensemble Learning
文章编号:
1001-2257 ( 2024 ) 11-0003-08
作者:
沈佳眉 1 ; 2 ; 朱 红 1
1. 国网江苏省电力有限公司南京供电分公司,江苏 南京 210019 ;
2. 南京大学计算机科学与技术系,江苏 南京 210023
Author(s):
SHEN Jiamei1 ;  2 ;  ZHU Hong1
( 1.Nanjing Power Supply Branch , State Grid Jiangsu Electric Power Co. , Ltd. , Nanjing 210019 , China ;
2.Department of Computer Science and Technology , Nanjing University , Nanjing 210023 , China )
关键词:
时间序列补全; 集成学习; 智能电表读数
Keywords:
time series imputation ;  ensemble learning ;  smart meter read data
分类号:
TM714
文献标志码:
A
摘要:
基于电力系统所采集的海量数据进行运行管控,在维持电网的安全、稳定和经济运行上发挥着重要的作用。由于设备故障和通信中断等原因,量测数据可能会出现缺失。因此,如何补全缺失数据成为电力系统研究的热点与关键问题。然而,单一的数据补全模型往往难以应对复杂多变的数据环境,如大段缺失和时间序列特征多样等,不同模型处理数据的能力各有优缺点。基于此,提出 AttnStack 算法,一个耦合注意力机制和集成学习的电力时间序列数据补全方法。该算法综合了先进模型图循环补全网络模型( GRIN )和时间序列双向循环补全模型( BRITS )的预测结果,引入时间断面的上下文信息,通过注意力机制在每一时间断面上动态调整每个个体学习器输出的权重。综合 GRIN 和 BRITS 算法使之具备同时处理剧烈波动和平缓稳定场景的能力。通过在来自爱尔兰的实际居民用户智能电表数据上进行实验,相比于仅使用单个模型,使用单一模型( 3 个 GRIN )的集成算法均方误差最少降低了 4.87% ,耦合 GRIN 和 BRITS 算法平均绝对误差降幅高达 19% ,证明 AttnStack 算法汇聚了各模型的优势,比单一模型更能应对大波动、少规律的用户负荷功率数据补全任务。
Abstract:
Leveraging the massive data collected from power systems for operational control plays a vital role in maintaining the safety , stability , and economic operation of the power grid.Due to equipment failures , communication interruptions , and other reasons , measurement data may be missing , making the imputation of missing data a hotspot and key issue in power system research.However , single data imputation models tend to struggle coping with complex and variable data environments , such as large missing segments , diverse time series features , etc.The ability of different models to deal with the data varies from one to another.In light of this , the AttnStack algorithm is proposed , a power time series data imputation method that couples attention mechanisms with ensemble learning.This algorithm integrates the imputation results of advanced models GRIN and BRITS , introduces the context information of time sections , and dynamically adjusts the output weight of each base learner on each time section through the attention mechanism.Combining the GRIN and BRITS models enables the handling of scenarios with both severe fluctuations and smooth stability.Experiments on real residential user smart meter data from Ireland show that comparing to use only one modle , using an ensemble algorithm of single models ( three GRINs ) reduces the? mean squared error at? least 4.87% , and coupling GRIN and BRITS algorithms reduces the mean absolute error by up to 19% , which demonstrates that , by converging the strengths of various models , AttnStack is more capable of handling the data imputation tasks of user load power data with large fluctuations and few regularity than a single model.

参考文献/References:

[ 1 ] 陈珩 . 电力系统稳态分析(第 4 版)[ M ] . 北京:中国电力出版社,2015.

[ 2 ] JIANG Z , SHI D , GUO X B , et al.Robust smart meter data analytics using smoothed ALS and dynamic time warping [ J ] .Energies , 2018 , 11 ( 6 ): 1401-1413.
[ 3 ] KUPPANNAGARI S R , FU Y , CHUENG C M , et al. Spatio-temporal missing data imputation for smart power grids [ C ] ∥Proceedings of the Twelfth ACM International Conference on Future Energy Systems. New York : ACM , 2021 : 458-465.
[ 4 ] PEPPANEN J , ZHANG X , GRIJALVA S , et al.Handling bad or missing smart meter data through advanced data imputation [ C ] ∥2016 IEEE Power and Energy Society Innovative Smart Grid Technologies Conference ( ISGT ) .New York : IEEE , 2016 : 1-5.
[ 5 ] FUNG D S.Methods for the estimation of missing values in time series [ D ] .Perth : Edith Cowan University , 2006.
[ 6 ] BORGES C E , KAMARA-ESTEBAN O , CASTILLO CALZADILLA T , et al.Enhancing the missing data imputation of primary substation load demand records [ J ] .Sustainable energy , grids and networks , 2020 , 23 : 100369.
[ 7 ] WU J J , KOIRALA A , VAN HERTEM D.Review of statistics based coping mechanisms for Smart Meter Missing Data in Distribution Systems [ C ] ∥2022 IEEE PES Innovative Smart Grid Technologies Conference Europe ( ISGT-Europe ) .New York : IEEE , 2022 : 1-6.
[ 8 ] PARK K , JEONG J , KIM D , et al.Missing-insensitive short-term load forecasting leveraging autoencoder and LSTM [ J ] .IEEE Access , 2020 , 8 : 206039-206048.
[ 9 ] RAGHUVAMSI Y , TEEPARTHI K.Distribution system state estimation with Transformer-Bi-LSTM based imputation model for missing measurements [ J ] . Neural computing and applications , 2024 , 36 ( 3 ):1295-1312.
[ 10 ] CAO W , WANG D , LI J , et al.Brits : bidirectional recurrent imputation for time series [ J ] .Advances in neural information processing systems , 2018 , 31 ( 1 ): 5159-5169.
[ 11 ] WU Z H , PAN S R , CHEN F W , et al.A comprehensive survey on graph neural networks [ J ] .IEEE Transactions on neural networks and learning systems , 2020 , 32 ( 1 ): 4-24.
[ 12 ] LIU Y X , JIN M , PAN S R , et al.Graph self-supervised learning : A survey [ J ] .IEEE Transactions on knowledge and data engineering , 2022 , 35 ( 6 ): 5879-5900.
[ 13 ] ANDREA C , IVAN M , ALIPPI C.Filling the G _ap_s : multivariate time series imputation by graph neural networks [ C ] ∥The Tenth International Conference on Learning Representations ( ICLR 2022 Poster ), 2022.
[ 14 ] KUMAR V , AYDAV P S S , MINZ S.Multi-view ensemble learning using multi-objective particle swarm optimization for high dimensional data classification [ J ] .Journal of King Saud University-computer and information sciences , 2022 , 34 ( 10 ): 8523-8537.
[ 15 ] MOHAMMED A , KORA R.A comprehensive review on ensemble deep learning : opportunities and challenges [ J ] .Journal of King Saud University-computer and information sciences , 2023 , 35 ( 2 ): 757-774.
[ 16 ] WOLPERT D H.Stacked generalization [ J ] .Neural networks , 1992 , 5 ( 2 ): 241-259.
[ 17 ] LIU W F , POKHAREL P P , PRINCIPE J C.Correntropy : properties and applications in non-Gaussian signal processing [ J ] .IEEE Transactions on signal processing , 2007 , 55 ( 11 ): 5286-5298.

备注/Memo

备注/Memo:
收稿日期: 2024-04-25
基金项目:国家重点研发计划( 2022YFB2404205 );国网江苏省电力有限公司科技项目( J2023066 )
作者简介:沈佳眉 ( 2000- ),女,浙江杭州人,硕士,助理工程师,研究方向为时间序列补全、预测技术;朱 红 ( 1971- ),女,江苏南京人,硕士,教授级高级工程师,研究方向为电力人工智能。
更新日期/Last Update: 2024-12-09