摘要
针对风电机组异常状态的检测问题,提出了考虑相似机组运行状态的风电机组异常检测方法。基于滑动时窗和K-means聚类算法对风电机组运行数据进行分析,提出了风电机组动态集群方法,进而建立了考虑时空相关性的风电机组集群。提出基于自适应权重与Levy飞行策略的北方苍鹰优化(WLNGO)算法;利用五折交叉验证(5CV)改进WLNGO算法,得到WLNGO-5CV算法,并利用该算法对核极限学习机(KELM)的超参数进行优化,进一步提出WLNGO-5CV-KELM回归模型。结合滑动时窗对相似机组预测残差进行统计分析得到实时预警阈值,消除了工况等因素对风电机组的影响,能够对目标风电机组进行可靠的异常检测。通过对中国东北某风电场的实际数据进行仿真分析,验证了所提方法的有效性和准确性。
Aiming at the problem of abnormal state detection of wind turbine(WT),a method of abnormal state detection of WT considering the operating state of similar WTs is proposed.Based on sliding timewindow and K-means clustering algorithm,the WT operation data are analyzed,the dynamic clustering method of WTs is proposed,and then the WT clustering considering spatio-temporal correlation is established.The WLNGO(adaptive weight and Levy flight based northern goshawk optimization)algorithm is proposed.The 5-fold cross validation(5CV)is used to improve WLNGO algorithm,and the WLNGO-5CV algorithm is pro-posed,which is used to optimize the hyperparameters of the kernel extreme learning machine(KELM),and the WLNGO-5CV-KELM regression model is further proposed.Then,the real-time warning threshold is obtained by combining the sliding time-window with the statistical analysis of the prediction residuals of similar WTs,which eliminates the influence of working conditions and other factors on WTs,and is able to reliably detect the anomalies of target WT.The effectiveness and accuracy of the proposed method are veri-fied by simulation analysis of actual data from a wind farm in Northeast China.
作者
于华楠
李靖雨
王鹤
李石强
边竞
YU Huanan;LI Jingyu;WANG He;LI Shiqiang;BIAN Jing(Key Laboratory of Modern Power System Simulation and Control&Renewable Energy Technology Ministry of Education,Northeast Electric Power University,Jilin 132012,China)
出处
《电力自动化设备》
北大核心
2025年第3期64-71,94,共9页
Electric Power Automation Equipment
基金
吉林省科技发展计划项目(20220203163SF)。
