Diagnostic Method for Load Deviation in Ultra-Supercritical Units Based on MLNaNBDOS  

作　　者：Mingzhu Tang YujieHuang Dongxu Ji Hao Yu 

机构地区：[1]School of Energy and Power Engineering,Changsha University of Science&Technology,Changsha,410114,China [2]School of Science and Engineering,The Chinese University of Hong Kong,Shenzhen,518000,China

出　　处：《Frontiers in Heat and Mass Transfer》2025年第1期95-129,共35页热量和质量传递前沿(英文)

基　　金：supported by the National Natural Science Foundation of China(Grant No.62173050);Shenzhen Municipal Science and Technology Innovation Committee(Grant No.KCXFZ20211020165004006);Natural Science Foundation of Hunan Province of China(Grant No.2023JJ30051);Hunan Provincial Graduate Student Research Innovation Project(Grant No.QL20230214);Major Scientific and Technological Innovation Platform Project of Hunan Province(2024JC1003);Hunan Provincial University Students’Energy Conservation and Emission Reduction Innovation and Entrepreneurship Education Center(Grant No.2019-10).

摘　　要：Load deviations between the output of ultra-supercritical(USC)coal-fired power units and automatic generation control(AGC)commands can adversely affect the safe and stable operation of these units and grid load dispatching.Data-driven diagnostic methods often fail to account for the imbalanced distribution of data samples,leading to reduced classification performance in diagnosing load deviations in USC units.To address the class imbalance issue in USC load deviation datasets,this study proposes a diagnostic method based on the multi-label natural neighbor boundary oversampling technique(MLNaNBDOS).The method is articulated in three phases.Initially,the traditional binary oversampling strategy is improved by constructing a binary multi-label relationship for the load deviations in coal-fired units.Subsequently,an adaptive adjustment of the oversampling factor is implemented to determine the oversampling weight for each sample class.Finally,the generation of new instances is refined by dynamically evaluating the similarity between new cases and natural neighbors through a random factor,ensuring precise control over the instance generation process.In comparisons with nine benchmark methods across three imbalanced USC load deviation datasets,the proposed method demonstrates superior performance on several key evaluation metrics,including Micro-F1,Micro-G-mean,and Hamming Loss,with average values of 0.8497,0.9150,and 0.1503,respectively.These results substantiate the effectiveness of the proposed method in accurately diagnosing the sources of load deviations in USC units.

关 键 词：Ultra-supercritical units load deviation multi-label learning class imbalance data oversampling 

分 类 号：G63[文化科学—教育学]