Suzhou Electric Appliance Research Institute
期刊號： CN32-1800/TM| ISSN1007-3175

SUBSCRIPTION MANAGEMENT

發(fā)行征訂

首頁 >> 發(fā)行征訂 >> 征訂方式

基于小波變換結(jié)合堆疊融合算法的非侵入式負(fù)載識別

來源：電工電氣發(fā)布時間：2025-10-28 15:28瀏覽次數(shù)：4

基于小波變換結(jié)合堆疊融合算法的非侵入式負(fù)載識別

李港，邱達(dá)，劉西林
(湖北民族大學(xué) 智能科學(xué)與工程學(xué)院，湖北 恩施 445000)
 
    摘 要：針對非侵入式負(fù)載監(jiān)測識別準(zhǔn)確率低、泛化能力弱、穩(wěn)定性差的問題，提出了一種結(jié)合特征選擇性小波變換與堆疊融合分類算法的負(fù)載識別方法。研究利用 CS5463 芯片采集電能數(shù)據(jù)，通過特征選擇性小波變換提取電流的時頻特征，并結(jié)合功率和功率因數(shù)構(gòu)建復(fù)合特征向量。采用k 最近鄰算法(KNN)、隨機(jī)森林(RF)和支持向量機(jī)(SVM)作為基學(xué)習(xí)器，通過堆疊融合算法提升準(zhǔn)確率、泛化能力，優(yōu)化分類性能，并引入動態(tài)負(fù)載識別優(yōu)化算法以提升實(shí)際應(yīng)用效果。實(shí)驗(yàn)結(jié)果表明，該堆疊融合模型在測試集上的準(zhǔn)確率為98.42%，而單一模型KNN、SVM和RF的準(zhǔn)確率分別為90.24%、94.99% 和97.10%，同樣數(shù)據(jù)集上未經(jīng)小波變換的融合算法準(zhǔn)確率為93.67%，加入動態(tài)負(fù)載識別優(yōu)化算法后，模型的穩(wěn)定性和準(zhǔn)確性在實(shí)際應(yīng)用中進(jìn)一步提高。
    關(guān)鍵詞： 非侵入式負(fù)載監(jiān)測；特征選擇性小波變換；堆疊融合算法；CS5463 芯片；動態(tài)負(fù)載識別優(yōu)化算法
    中圖分類號：TM714 ；TM734     文獻(xiàn)標(biāo)識碼：A     文章編號：1007-3175(2025)10-0031-07
 
A Non-Intrusive Load Identification Method Based on Wavelet
Transform and Stacked Fusion Algorithm
 
LI Gang, QIU Da, LIU Xi-lin
（College of Intelligent Systems Science and Engineering, Hubei Minzu University, Enshi 445000, China）
 
    Abstract: To address the challenges of low identification accuracy, weak generalization capability, and poor stability in non-intrusive load monitoring,this paper proposes a load identification method that integrates feature-selective wavelet transform with a stacked fusion algorithm. The study utilizes the CS5463 chip to collect electrical data, extracts the time-frequency characteristics of current signals by applying feature-selective wavelet transform, and combines with power and power factor information to construct a composite feature vector. Subsequently, k-nearest neighbors (KNN) algorithm, random forests (RF) , and support vector machines (SVM) are employed as base learners, the accuracy and generalization ability are enhanced through the stacked fusion algorithm, the classification performance is optimized, and the dynamic load identification optimization algorithm is introduced to improve the practical application effect. Experimental results demonstrate that the accuracy rate of the stacked fusion model on the test set is 98.42%, while the accuracy rates of the single models KNN, SVM and RF are 90.24%, 94.99% and 97.10% respectively. The accuracy rate of the fusion algorithm without wavelet transform on the same dataset is 93.67%. After adding the dynamic load identification optimization algorithm,the stability and accuracy of the model have been further enhanced in practical applications.
    Key words: non-intrusive load monitoring; feature-selective wavelet transform; stacked fusion algorithm; CS5463 chip; dynamic load identification optimization algorithm
 
參考文獻(xiàn)
[1] 陳繼開，祝世啟，李浩茹，等. 弱電網(wǎng)下并網(wǎng)逆變器鎖相環(huán)優(yōu)化方法[J]. 儀器儀表學(xué)報(bào)，2022，43(2) ：234-243.
[2] REHMAN A U, TITO S R, NIEUWOUDT P, et al.Applications of Non-Intrusive Load Monitoring Towards Smart and Sustainable Power Grids:A System Perspective[C]//2019 29th Australasian Universities Power Engineering Conference(AUPEC)，2019 ：1-6.
[3] HART G W.Nonintrusive appliance load monitoring [J].Proceedings of the IEEE，1992，80(12) ：1870-1891.
[4] ZEIFMAN M, ROTH K.Nonintrusive appliance load monitoring: Review and outlook[J].IEEE Transactions on Consumer Electronics，2011，57(1) ：76-84.
[5] NGUYEN M, ALSHAREEF S, GILANI A, et al.A novel feature extraction and classification algorithm based on power components using single-point monitoring for NILM[C]//2015 IEEE 28th Canadian Conference on Electrical and Computer Engineering(CCECE)，2015 ：37-40.
[6] LU Lei, GU Chao, FENG Junguo, et al.Non-Intrusive Load Monitoring Based on Multiple Feature Optimization and Genetic Algorithm[C]//2022 5th International Conference on Renewable Energy and Power Engineering(REPE)，2022 ：115-120.
[7] SUN Mingxu, NAKOTY Francis Mawuli, LIU Qi, et al.Non-intrusive load monitoring system framework and load disaggregation algorithms:A survey[C]//2019 International Conference on Advanced Mechatronic Systems(ICAMechS)，2019 ：284-288.
[8] PRECIOSO D, GOMEZ-ULLATE D.Thresholding methods in non-intrusive load monitoring[J].The Journal of Supercomputing，2023，79(13) ：14039-14062.
[9] MOHAMMAD I A, RAJABI R, ESTEBSARI A.Non-Intrusive Load Monitoring(NILM) Using Deep Neural Networks:A Review[EB/OL].(2023-06-08)[2025-06-05].https://arxiv.org/pdf/2306.05017.
[10] MURSHED M G S, MURPHY C, HOU D, et al.Machine learning at the network edge:A survey[J].ACM Computing Surveys(CSUR)，2021，54(8) ：1-37.
[11] GUIDOTTI R, MONREALE A, RUGGIERI S, et al.A survey of methods for explaining black box models[J].ACM Computing Surveys(CSUR)，2018，51(5) ：1-42.
[12] 董慧芬，陳蒙. 電能質(zhì)量信號的非均勻子帶分解小波去噪[J]. 電子測量與儀器學(xué)報(bào)，2022，36(3) ：149-156.
[13] HE Hui, LIU Zixuan, JIAO Runhai, et al.A novel nonintrusive load monitoring approach based on linear-chain conditional random fields[J].Energies，2019，12(9) ：1797.
[14] CHEN Junfeng, WANG Xue, ZHANG Xiaotian, et al.Temporal and spectral feature learning with twostream convolutional neural networks for appliance recognition in NILM[J].IEEE Transactions on Smart Grid，2022，13(1) ：762-772.
[15] FANG Zhaoyuan, ZHAO Dongbo, CHEN Chen, et al.Nonintrusive appliance identification with appliance-specific networks[J].IEEE Transactions on Industry Applications，2020，56(4) ：3443-3452.
[16] GHOSH S , CHATTERJEE A , CHATTERJEE D . An improved load feature extraction technique for smart homes using fuzzy-based NILM[J].IEEE Transactions on Instrumentation and Measurement，2021，70 ：1-9.
[17] REHMAN A U , LIE T T , VALLES B , et al .Comparative evaluation of machine learning models and input feature space for non-intrusive load monitoring[J].Journal of Modern Power Systems and Clean Energy，2021，9(5) ：1161-1171.
[18] SAHA D, BHATTACHARJEE A, CHOWDHURY D, et al.Comprehensive NILM framework:Device type classification and device activity status monitoring using capsule network[J].IEEE Access，2020，8 ：179995-180009.
[19] 朱浩，曹寧，鹿浩，等. 基于特征加權(quán) KNN 的非侵入式負(fù)荷識別方法[J]. 電子測量技術(shù)，2022，45(8) ：70-75.
[20] LI Ding, DICK Scott.Residential household nonintrusive load monitoring via graph-based multilabel semi-supervised learning[J].IEEE Transactions on Smart Grid，2019，10(4) ：4615-4627.
[21] BERGES M, GOLDMAN E, MATTHEWS H S, et al. Training load monitoring algorithms on highly sub-metered home electricity consumption data[J].Tsinghua Science and Technology，2008，13(S1) ：406-411.
[22] CHEN M T, LIN C M.Standby power management of a smart home appliance by using energy saving system with active loading feature identification[J].IEEE Transactions on Consumer Electronics，2019，65(1) ：11-17.
[23] LIU Yanchi, WANG Xue, YOU Wei.Non-intrusive load monitoring by voltage-current trajectory enabled transfer learning[J].IEEE Transactions on Smart Grid，2019，10(5) ：5609-5619.