复杂环境中SF6气体泄漏红外图像监测算法研究  

Research on infrared image monitoring algorithms for SF_(6)gas leakage in complex environments

作　　者：王振兴[1] 肖光宇 刁目鑫 任志刚 石磊 WANG Zhenxing;XIAO Guangyu;DIAO Muxin;REN Zhigang;SHI Lei(State Key Laboratory of Electrical Insulation and Power Equipment,Xi’an Jiaotong University,Xi’an 710049,China;Beijing Electric Power Research Institute,Beijing 100075,China)

机构地区：[1]电工材料电气绝缘全国重点实验室,西安交通大学,陕西西安710049 [2]国网北京电科院,北京100075

出　　处：《电工电能新技术》2023年第11期93-104,共12页Advanced Technology of Electrical Engineering and Energy

基　　金：国家自然科学基金项目(51937009)。

摘　　要：SF6气体因其优异的绝缘和灭弧性能被广泛应用于各类电力设备中,对电力设备中的SF6泄漏的监测及预警研究,不但对电力系统的安全稳定运行具有重大意义,还是建设生态文明实现碳中和的有力举措。为解决实际工程中出现的多种因素对气体泄漏监测的干扰,本文针对复杂环境建立了SF6气体泄漏实验平台,获得不同条件下的气体泄漏红外图像数据集,并结合红外数据增强的Yolov4算法实现了对SF6气体泄漏监测。实验结果表明,在复杂环境中本文提出的监测模型对SF6气体泄漏的监测精度为80%,相对于当下主流目标监测模型,表现出更好的鲁棒性。同时,将Yolov4算法与图像增强算法结合,可提升识别精度约9.08%,优化了模型稳定性。然后分析在同一红外成像视角下发生多处SF6泄漏的监测结果,结果显示多点泄漏的识别精度低于单点泄漏,但在工程中依旧具有实用性。最后对比红外相机不同安放位置对于气体泄漏识别的影响,发现当相机与泄漏点垂直时(θ=90°),本模型的识别置信度可达到0.95,而当相机处于泄漏点侧面时(θ=10°),置信度降为0.51,但该精度依然满足实际监测需要。SF_(6)is widely used in all kinds of power equipment because of its excellent insulation and arc extinguishing properties.The monitoring and early warning research of SF_(6)leakage in power equipment is not only of great significance to the safe and stable operation of power system,but also a powerful initiative to build ecological civilization and realize carbon neutrality.In order to solve the problems of influence of complex environment on gas leakage monitoring,this paper establishes an experimental platform for SF_(6)leakage in complex environment,obtains infrared image data sets of gas leakage under different conditions,and realizes SF_(6)leakage monitoring by combining infrared data enhancement with Yolov4 algorithm.The experimental results show that the monitoring accuracy of the proposed monitoring model for SF_(6)leakage in complex environment is 80%,which shows better robustness as compared with the mainstream target monitoring model nowadays.Meanwhile,the combination of Yolov4 algorithm and image enhancement algorithm can improve the recognition accuracy by about 9.08%and optimize the model stability.Then analyze the monitoring results of multiple SF_(6)leaks occurring under the same infrared imaging view,the results show that the recognition accuracy of multi-point leaks is lower than that of single-point leaks,but it is still practical in engineering.Finally,comparing the effects of different placement of infrared cameras on gas leak identification,it is found that when the camera is perpendicular to the leak(θ=90°),the identification confidence of the model can reach 0.95,while when the camera is on the side of the leak(θ=10°),the confidence drops to 0.51,but the accuracy still meets the actual monitoring needs.

关 键 词：SF6气体泄漏 红外成像 图像识别 复杂环境 Yolov4 

分 类 号：TM595[电气工程—电器]