基于机器学习的电弧故障检测算法研究

发布时间：2017-12-28 16:29

本文关键词：基于机器学习的电弧故障检测算法研究　出处：《杭州电子科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：电弧故障是当前引起电气火灾的重要原因之一,电弧故障检测装置就能及时发现电弧故障并断开回路,从而避免火灾的发生。因此,电弧故障检测技术成为当前防火灾技术领域的一个研究热点。本文提出了一个基于机器学习的电弧故障检测算法总体框架,包括电弧故障特征提取、数据预处理和电弧故障分类器三部分,在算法训练时,从电弧故障样本数据中提取电弧故障特征参数并进行预处理,对电弧故障分类器进行训练,建立分类器模型;分类器模型建立之后,算法就可以进行实际电弧故障检测,从实测电路数据中提取电弧故障特征参数并进行预处理,采用训练建立分类器模型分类出其中的电弧故障数据,从而检测出电弧故障的发生。论文对电弧故障数据进行时域、频域和层次分析,确定了电弧故障检测算法需要提取的能量、子带能量比、短时平均能量、频谱质心、带宽、过零率、脉冲个数、方差、电流积分等参数电弧故障数据特征参数。论文结合电弧故障检测的特点,设计了电弧故障检测算法中的数据预处理方法,包括基于压缩近邻规则算法的数据清洗、基于特征标准化的数据归一化和基于主成分分析PCA算法的数据降维。论文为训练出泛化能力足够强的电弧故障分类器,提出了一个Logistic回归、SVM和随机森林等三种分类器相结合的组合分类器,并设计和实现了三种分类器以及第二层分类器。对电弧故障检测算法的测试表明,算法在电弧故障检测的电弧故障识别率和误动作两个主要性能指标上均获得了比较满意的效果。
[Abstract]:Arc fault is one of the most important causes of electrical fire. Arc fault detection device can detect arc faults and break the circuit in time to avoid fire. Therefore, the arc fault detection technology has become a research hotspot in the field of fire prevention and disaster technology. This paper presents an arc fault detection based on machine learning algorithm framework, including extraction, arc fault feature data preprocessing and arc fault classifier training algorithm in the three part, the extraction of fault feature parameters from the arc arc fault sample data and pretreatment, training of arc fault classifier, establish classification model; after the classifier model, the algorithm can extract the actual arc fault detection, arc fault characteristic parameters from the measured circuit data and pretreatment, the establishment of training classifier model classification of arc fault data of them, so as to detect the occurrence of arc fault. The paper carries on the analysis of time and frequency and level of arc fault data, determine the arc fault detection algorithm to extract energy and sub band energy ratio, short-time average energy, spectral centroid, bandwidth, zero crossing rate, pulse number, variance, the current integral parameters such as arc fault characteristic parameter. Based on the characteristics of arc fault detection, a data preprocessing method is designed for arc fault detection algorithm, including data cleaning based on compressed neighbor rule algorithm, normalization based on feature normalization and data reduction based on principal component analysis PCA algorithm. In order to train the arc fault classifier with strong generalization ability, this paper proposes a combination classifier based on three classifiers, which are Logistic regression, SVM and random forest, and designs and implements three classifiers and second level classifiers. The test of arc fault detection algorithm shows that the algorithm achieves satisfactory results in the two main performance indexes of arc fault recognition rate and misoperation in arc fault detection.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TM501.2

【相似文献】