基于多信息融合的低压交流故障电弧早期征兆及高精度诊断算法研究

With the continuous improvement of the degree of electrification & intelligence and the sharp rise in electricity consumption in modern society, people's lives and property security is seriously threatened by electrical fire. Arc-fault is the first, the main and the important cause of electrical fire. At present, there are some problems in the detection of arc-fault such as the feature information of the arc-fault can not be accurately extracted, the protection method for it is passive which alarming after the arc has burning, etc. This project takes the arc-fault of low voltage power distribution system as the research object, the main line of the project is to search the early warning signs which is easy to be confused with the normal working arc and the high precision identification algorithm, the goal of the project is to comprehensive improve the rapidity, accuracy and high accuracy of the identification of arc-fault. Based on the experience of the pre-built arc-fault simulation test platform, further build the single/three phase arc-fault simulation test platform which meet the latest international UL1699 (Third Edition) standards and combined with the characteristics of China's power grid, experimental study the early arcing symptoms of arc-fault at the case of a variety of typical load, special load, combined load and disturbance load, to explore a more reasonable of the combination form of the characteristics signal and signal denoising method, propose a new arc-fault fusion algorithm based on adaptive weighting & neural network, provide the theory and enabling technology support for the realization of the early warning of arc-fault & the improvement the accuracy rate of arc-fault alarm.

随着现代社会电气化、智能化程度的不断提高，用电量的急剧攀升，电气火灾对人民生命财产安全带来严重威胁。故障电弧是引起电气火灾的首要、主要和重要原因。目前对故障电弧的检测存在不能准确提取其特征信息及对其的保护采用被动的燃弧后的报警机制等问题。本项目以低压配电系统故障电弧为研究对象，以寻找易于与正常工作电弧混淆的故障电弧早期燃烧征兆、高精度辨识算法为主线，以综合提高故障电弧辨识的快速性、准确性及高精度为目标，在前期搭建故障电弧模拟试验平台的经验基础上，进一步搭建符合国际UL1699(第三版)最新标准并结合我国电网特征的单/三相故障电弧模拟试验平台，试验研究多种典型负载、特殊负载、组合负载以及干扰负载情况下故障电弧早期燃弧征兆，探索更为合理的特征信号组合形式及信号消噪方法，提出一种新的基于自适应加权结合神经网络的故障电弧融合算法，为实现故障电弧早期预警、提高故障电弧报警准确率提供理论和使能技术支撑

本项目“基于多信息融合的低压交流故障电弧早期征兆及高精度诊断算法研究”主要围绕故障电弧高精度检测进行研究。故障电弧是引起电气火灾的首要、主要和重要原因，又是目前还没有完全攻克的难点。本项目旨在故障电弧的高精度检测上进行深度的创新性研究，以期获得对故障电弧很好的检出方法，为故障电弧的检测、我国电气火灾报警系统的完善、我国智慧消防事业的建设贡献一定的力量。消防事业是国家长治久安、人民幸福生活，保障国民经济社会健康发展的重要基石，而故障电弧的检测是发展消防事业的短板，也是一个突破口，若能够安全、有效、高精度地在线检测出故障电弧的发生，必将对我们消防系统的完善起到很大的推动作用，也会给整个行业的构架带来新的改变和新的机遇。. 本项目的研究内容主要有：研发出一套具有多功能性的符合国际UL1699标准并结合我国电网特性的故障电弧模拟试验平台；进行了基于AlexNet深度学习网络的串联故障电弧检测研究，随后又提出了基于ResNet深度学习网络的串联故障电弧检测研究，将单一负载的检测准确率提高到99.75%，将整体检测准确率提高到97.125%。提出基于小波变换及深度神经网络的故障电弧检测方法研究，测试结果表明基于小波特征深度神经网络方法具有较高的故障辨识准确率，为99.19%，误差率较低情况下召回率同样相对较高。另该模型具有一定的泛化能力，具有一定的通用性。随后又进行了故障电弧负载类型分类辨识的实验与研究，结果表明在区分不同负载的情况下总体准确率达到93.03%。考虑到故障电弧采集数据具有时间上的前后连续特性，提出一种基于LSTM长短期记忆网络的故障电弧检测方法。并针对现实情况，提出开展干/支路故障电弧检测研究，即模拟最具现实情况的每一个用电电器作为一个用电支路，其上发生故障电弧，将检测点设置在干路上，即监测干路电流实现对支路故障电弧的预测预警。该项研究目前正在稳步推进，预期能够获得比较好的研究结果。