电力系统及其自动化学报2025，Vol.37Issue(2)：48-57,10.DOI:10.19635/j.cnki.csu-epsa.001480
基于增量学习的光伏并网逆变器故障诊断
Fault Diagnosis of Photovoltaic Grid-connected Inverter Based on Incremental Learning
摘要
Abstract
The existing fault diagnosis techniques for photovoltaic(PV)grid-connected inverters based on offline learn-ing require retraining the model on the entire dataset when updating the model parameters,which poses challenges in identifying new types of faults and lacks model flexibility.To address these shortcomings,a fault diagnosis method for PV grid-connected inverters is proposed by combining multi-scale morphology and incremental learning.First,the three-phase current fault signals are processed through multi-scale morphological filtering,followed by segmenting the processed signals using a sliding window to obtain the fault dataset which is divided into a historical dataset and a new dataset.Then,a one-dimensional convolutional neural network is used to learn from the historical data,and a nearest-mean-of-exemplars classifier is employed to identify the types of historical faults.Finally,a herding algorithm is used to construct representative samples,and distillation loss is added to the original loss function to retrain the model,thus achieving the identification of new fault types.Simulation results demonstrate that the proposed method can effectively distinguish between new and historical fault types and overcome catastrophic forgetting.In addition,the model exhibits a high accuracy and a high robustness.关键词
光伏并网逆变器/增量学习/数学形态学/故障诊断/卷积神经网络Key words
photovoltaic(PV)grid-connected inverter/incremental learning/mathematical morphology/fault diagno-sis/convolutional neural network分类
信息技术与安全科学引用本文复制引用
公铮,丁家伟,刘允浩,李武能..基于增量学习的光伏并网逆变器故障诊断[J].电力系统及其自动化学报,2025,37(2):48-57,10.基金项目
国家自然科学基金资助项目(52277205) （52277205）
江苏省自然科学基金资助项目(BK20230108). （BK20230108）
