基于有监督机器学习的光伏故障监测系统研究进展  

Review of Supervised Machine Learning Based on Photovoltaic Fault Monitoring System

作　　者：董煌锋 郭信平 冀璇 肖文波[1,2] DONG Huangfeng;GUO Xingping;JI Xuan;XIAO Wenbo(Key Laboratory of Nondestructive Testing Technology of Ministry of Education,Nanchang Hangkong University,Nanchang 330063,China;Engineering Laboratory of Optoelectronics Detection Technology in Jiangxi Province,Nanchang Hangkong University,Nanchang 330063,China)

机构地区：[1]南昌航空大学无损检测技术教育部重点实验室,江西南昌330063 [2]南昌航空大学江西省光电检测技术工程实验室,江西南昌330063

出　　处：《成都大学学报（自然科学版）》2024年第3期262-273,共12页Journal of Chengdu University（Natural Science Edition）

基　　金：国家自然科学基金(12064027、62065014);研究生创新专项(YC2022-113、YC2022-118)。

摘　　要：系统性总结了有监督机器学习在光伏故障监测技术中的应用.支持向量机(SVM)对惩罚因子和核函数非常敏感,通过优化参数选择和数据预处理可以提高监测准确率.决策树(DT)容易过拟合,可以通过剪枝技术避免过拟合问题.随机森林(RF)对数据量和参数调节要求较高,可以通过算法生成数据和优化参数来满足要求,从而提高监测准确率.K-近邻(KNN)在处理高维数据时能力较差,可以引入合适的核函数和数据预处理技术来提高准确率.神经网络(ANN)需要大量数据和参数选择,优化算法可以解决这些问题.ANN和SVM具有最高的准确率但耗时较长,DT耗时短但准确率较低.未来的趋势是进一步优化算法,结合深度学习和智能化发展.由于ANN在故障监测中具有高准确率,基于ANN的光伏故障监测系统预计将成为主流方法.This article provides a systematic summary of supervised machine learning techniques for photovoltaic fault monitoring.Support Vector Machine(SVM)is sensitive to the penalty factor and kernel function,and its accuracy can be improved by optimizing parameter selection and preprocessing the data.Decision Tree(DT)is prone to overfitting,but this can be avoided by using pruning technique.Random Forest(RF)requires high-quality data and tuning,which can be achieved by generating data through algorithms and by using optimization algorithms for parameter tuning,thereby improving the monitoring accuracy.K-Nearest Neighbors(KNN)has poor performance in handling high-dimensional data,but an appropriate kernel function can be introduced to address this issue,and data preprocessing techniques can also improve KNN s monitoring accuracy.Artificial Neural Network(ANN)requires a large amount of data for training and parameter selection,which can be addressed by optimization algorithms.Furthermore,ANN and SVM have the highest accuracy,albeit time-consuming,whereas DT is faster,but less accurate.The future trend is to further optimize the algorithms,integrate them with deep learning,and develop intelligent systems.ANN s high accuracy in fault monitoring,and the prediction systems based on neural networks are expected to become the mainstream method of fault monitoring.

关 键 词：光伏故障 有监督机器学习 神经网络 监测系统 

分 类 号：TM615[电气工程—电力系统及自动化] TP274[自动化与计算机技术—检测技术与自动化装置]