摘要
发电厂在变负荷工况下的短路电流具有显著的时变与非线性特性,传统方法难以对其动态过程进行准确跟踪。为此,本文提出一种融合多尺度卷积与时序注意力机制的神经网络系统。该系统利用多尺度卷积提取电流信号的多频带特征,并通过时序注意力机制增强关键故障时段的信息表达能力,从而实现对短路电流的高精度测量。实验结果表明,在轻载、满载和过载工况下,系统故障识别准确率分别达到97.5%、96.9%和95.8%,平均响应时间低于10.5 ms。此外,在不同气象条件与电网扰动下,系统仍表现出良好的一致性与鲁棒性。本研究为发电厂短路故障电流的智能感知与运维提供了有效技术支撑,对提升电力系统安全稳定运行水平具有重要意义。
Short-circuit currents in power plants under varying load conditions exhibit significant time-varying and nonlinear characteristics,making it difficult for traditional methods to accurately track their dynamic processes.Therefore,this paper proposes a neural network system integrating multi-scale convolution and temporal attention mechanisms.This system utilizes multi-scale convolution to extract multi-band features of the current signal and enhances the information representation capability during critical fault periods through a temporal attention mechanism,thereby achieving high-precision measurement of short-circuit currents.Experimental results show that the system achieves fault identification accuracy of 97.5%,96.9%,and 95.8%under light load,full load,and overload conditions,respectively,with an average response time of less than 10.5 ms.Furthermore,the system maintains good consistency and robustness under different meteorological conditions and grid disturbances.This research provides effective technical support for the intelligent sensing and operation and maintenance of short-circuit fault currents in power plants,which is of great significance for improving the safe and stable operation of power systems.
作者
杨倩
孙元坤
高佳兴
杨凡
蔡昊晨
黄稳
YANG Qian;SUN Yuankun;GAO Jiaxing;YANG Fan;CAI Haochen;HUANG Wen(CHN Energy Shenhua Jiujiang Power Generation Co.,Ltd.,Jiujiang 332000,China;Anhui Zhongke Heneng Power Technology Co.,Ltd.,Hefei 230088,China)
出处
《国外电子测量技术》
2025年第11期164-169,共6页
Foreign Electronic Measurement Technology
基金
国家能源集团科技创新项目(SHJJ-26-KJ-01)。
关键词
短路故障
多尺度卷积
时序注意力机制
故障识别
short-circuit fault
multi-scale convolution
temporal attention mechanism
fault identification
