基于改进残差网络的光伏逆变器数据异常检测方法  

DATA ANOMALY DETECTION METHOD FOR PHOTOVOLTAIC INVERTER DATA BASED ON IMPROVED RESIDUAL NETWORK

作　　者：姚森山 庞成鑫 Yao Senshan;Pang Chengxin(School of Electronics and Information Engineering,Shanghai University of Electric Power,Shanghai 201306,China)

机构地区：[1]上海电力大学电子与信息工程学院,上海201306

出　　处：《太阳能学报》2025年第4期322-330,共9页Acta Energiae Solaris Sinica

基　　金：国家电网有限公司科学技术项目(SGSCJY00GHJS2000014)。

摘　　要：为解决光伏组件异常导致光伏逆变器数据异常的问题,提出一种基于改进残差网络(SCCB-ResNet50)的光伏逆变器数据异常检测方法。该方法引入马尔可夫转移场将光伏功率时序数据化转为二维图像,以增加数据特征点从而提高检测精度,同时使用改进残差网络提取数据异常特征,进行数据异常检测。改进残差网络在残差网络中引入通道注意力和空间注意力融合机制(CBAM),并使用改进的随机梯度下降(SGD)优化器和余弦退火学习率下降策略,以提高数据异常检测精度。结果表明:该方法在AUC、召回率和准确率上分别达到95.8%、81.5%、96.0%,与LSTM等其他数据异常检测方法相比,3个评价指标均得到有效提高,具有优异的数据异常检测能力。In order to solve the problem of PV module anomalies leading to PV inverter data anomalies,a PV inverter data anomaly detection method based on improved residual network(SCCB-ResNet50)is proposed.The method introduces a Markov transfer field to convert the PV power time series data into a two-dimensional image in order to increase the data feature points and thus improve the detection accuracy,and also extracts the data anomaly features using the improved residual network for data anomaly detection.The improved residual network introduces channel attention and spatial attention fusion mechanism in the residual network and uses improved SGD optimizer and cosine annealing learning rate reduction strategy to improve the data anomaly detection accuracy.The experimental results show that;the method achieves 95.8%,81.5%and 96.0%in AUC,recall rdtio,and accuracy,respectively.Compared with other data anomaly detection methods such as LSTM,all the three evaluation indexes are effectively improved and have excellent data anomaly detection capability.

关 键 词：异常检测 光伏逆变器 故障分析 机器学习 残差网络 马尔可夫过程 

分 类 号：TM615[电气工程—电力系统及自动化]