复杂组合负载条件下室内配电系统串联电弧故障诊断

The safety commission of the state council launched a three-year comprehensive management of electrical fires in 2017, due to frequent electrical fires in recent years. When a series arc fault occurs in indoor power distribution system, the current value of circuit is often less than the threshold of the circuit breaker. But the temperature of arc combustion can be as high as thousands of degrees, which can lead to electrical fire. At present, the study on series arc fault of indoor power distribution system is mainly aimed at single or simple combined load, while most indoor power distribution systems have complex combined loads. The physical properties of complex combined load circuits are different from those of single or simple combined load circuits. This project establishes an arc simulation model based on physical parameters and energy calculation, to explore the mechanism of series arc fault under complex combined loads conditions and clarify the impact of arc fault on physical quantities such as current and voltage. The self-coding method of deep learning key information is proposed to extract the key features of arc fault. A sparse representation method of adjustable parameters and multi-dictionary learning is proposed to detect series arc fault. The circuit parameters analysis method is proposed to provide a basis for arc fault location. A real-time diagnosis system of series arc fault is developed to detect and locate the series arc fault accurately.

近年来电气火灾高发，2017年国务院安委会启动为期三年的电气火灾综合治理工作。室内配电系统发生串联电弧故障时，电流一般较小且低于断路器阈值，但产生的温度可高达几千度且不易熄灭，成为导致电气火灾的主要原因之一。目前关于室内配电系统串联电弧故障的研究主要针对单一或简单组合负载，而实际室内配电系统多为复杂组合的负载。复杂组合负载电路中的物理量特征与单一或简单组合负载电路有所不同。本项目提出基于物理参数和能量计算的方法建立电弧仿真模型，探索复杂组合负载条件下室内配电系统串联电弧故障发生机理,明确发生电弧故障对电流、电压等物理量的影响。提出深度学习关键信息自编码方法，提取故障关键特征信息。提出可调参数多字典学习稀疏表示方法,实现电弧故障检测。提出基于电路参数分析的故障定位理论，为电弧故障定位提供依据。研制串联电弧故障实时诊断系统，实现复杂室内配电系统串联电弧故障的准确检测与精准定位。

根据相关资料分析，室内配电系统的串联电弧故障已成为电气火灾的主要原因之一。本项目研究不同条件下多种有效的串联电弧故障诊断理论与技术，实现串联电弧故障的准确检测与精确定位，对于预防和控制电气火灾具有重要的理论与实际意义。主要研究内容及重要结果包括：①通过仿真室内不同类型负载配电系统串联电弧故障发生机理，明确发生电弧故障对电流、电压等物理量的影响。②单相桥式整流电路电弧故障仿真，选取整流电路正常和故障情况在不同负载阻值下的奇数高次谐波幅值的基波百分比为特征值。③基于K均值聚类算法的电弧故障检测，利用K-means算法将原始数组分类为故障或正常信号并生成新的标签位。精度测试的原理是对比输入数据的标签位与算法计算后的数据标签位，若相同则返回1，不同则返回0。通过计算返回数据中0的比例计算每一个坐标系下的数据分类精度。经过精度测试后得出，70组数据在所有的坐标系下分类的精度均为100%。④基于卷积神经网络和长期短期记忆网络的局部放电识别算法，提出了基于卷积神经网络和长期短期记忆网络的 PD识别算法，并引入 Attention机制通过映射加权和学习参数矩阵赋予 LSTM隐含状态不同的权重，减少历史信息的丢失并加强重要信息的影响。建立电压信号变化与其电网运行状态响应之间的复杂关系。以 VSB大学公开的 ENET数据集对所提出的方法进行验证。提出的方法对正常类型和故障类型的预测精度分别达到了95.16%和 94.44%。并与多种算法进行对比，结果表明所提方法具有更高的识别精度。⑤基于小波压缩重构正则化的深度残差网络电弧故障检测，为了获得特征数据，对时域信号进行5层db4小波分析，将得到的分层小波系数映射到HSV颜色索引的相空间中制成图像数据集。针对图像数据集较少的问题，通过在一定范围内随机改变HSV的颜色索引值和随机H通道色相变化对原有数据集进行了扩增。⑥基于可调正则化阶次和多字典的稀疏表示方法实现电弧故障检测。