摘要
针对传统选线方法精度低、实时性差、易受噪声干扰的问题,提出一种混合数据驱动的轻量化YOLOv5选线方法,简记为MSE-YOLOv5。首先,以零序电流作为区分故障线路与非故障线路的判断依据,为了增强二者间数据对比差异性,利用小波变换将零序电流信号映射为二维时频图;其次,为了扩充样本数量,利用搭建的小电流接地系统仿真模型,通过改变故障点位置、初相位以及接地电阻等参数生成仿真数据,与真实数据构成混合数据集;然后,为了减少选线时背景噪声对微弱故障信号特征的影响,在所搭建检测网络的颈部网络中引入通道注意力模块,从而增强故障特征的表达能力;最后,为了提高选线实时性,在网络中引入轻量化网络以减少其参数量与运算量。为了验证所提出方法的优势,利用某变电站真实故障数据进行测试,并与4种经典算法进行比较。结果表明:所提混合数据驱动的轻量化YOLOv5故障选线方法具有较高精度,其选线精度可达95.2%,即使在噪声干扰条件下,选线精度依然可以保持在90%以上;具有更轻的体量及更快的选线速度,参数量下降至原网络的1/5,计算量下降至1/7,检测速度可达7.7 ms。因此,混合数据驱动的轻量化YOLOv5故障选线方法具有体量小、速度快、精度高的优点,有利于后期将其部署到现场设备中。
In order to solve the problems of low precision,poor real-time performance and being easy to be disturbed by noise in traditional fault line selection methods,a hybrid data-driven lightweight YOLOv5 line selection method is proposed,briefly referred to as MSE-YOLOv5.Firstly,the zero sequence current is used as the basis to distinguish the fault line from the non-fault line.To enhance the data difference between them,the zero sequence current signal is mapped to two-dimensional time-frequency graph by wavelet transform.Secondly,to expand the number of samples,the simulation model of the small current grounding system is used to generate the simulation data by changing the fault location,initial phase and fault resistances.Simulation data and real data constitute mixed data set together.Thirdly,to reduce the influence of background noise on weak fault signal characteristics during fault line selection,a channel attention module is introduced into the neck network to promote the expression ability of fault characteristics.Finally,to improve the real-time performance of fault line selection,lightweight network is introduced to reduce the number of parameters and calculation.In order to verify the advantages of the proposed method,the real fault data of a substation are tested and compared with those by four classical algorithms.The experimental results demonstrate that the proposed hybrid data-driven lightweight YOLOv5 fault line selection method exhibits a higher level of accuracy,with its line selection accuracy reaching 95.2%.Even in the presence of noise interference,the accuracy remains above 90%.Additionally,this method offers reduced weight and faster line selection speed,as it reduces the number of parameters to only 1/5 of the original network while decreasing computational requirements to just 1/7.Consequently,the detection speed can achieve an impressive rate of 7.7 ms.Therefore,the lightweight YOLOv5 line selection method driven by hybrid data has the advantages of compact size and high speed,suitable for deployment on field equipment in the future.
作者
郝帅
田卓
马旭
李威
李嘉豪
HAO Shuai;TIAN Zhuo;MA Xu;LI Wei;LI Jiahao(College of Electrical and Control Engineering,Xi’an University of Science and Technology,Xi’an 710054,China)
出处
《西安科技大学学报》
CAS
北大核心
2024年第5期966-975,共10页
Journal of Xi’an University of Science and Technology
基金
国家自然科学基金项目(51804250)
中国博士后科学基金项目(2020M683522)
陕西省自然科学基础研究计划项目(2024JC-YBMS-490)。
