摘要
根据绝缘子在线检测的特点,提出了基于改进ART2神经网络的绝缘子故障在线诊断技术.改进的ART2神经网络在F2层增设初始状态层,即F2层分为上子层F22和下子层F21.F22层存储绝缘子首次投运时的初始状态和已知的典型故障类型,F21层存储新增故障类型.在绝缘子故障诊断过程中,首先应用电晕电流脉冲测量仪在线测量绝缘子的电晕电流脉冲,然后绘制电晕电流脉冲的N-φ图,并将其作为故障诊断的特征量送入改进的ART2神经网络进行模式识别.被检测的特征量先与F22层中的初始状态和已知的典型故障类型进行比较,如果匹配则判为该类;如果不匹配,再与F21层中的新增故障类型进行比较;如果都不匹配,则在F21层中创建新增故障类型.改进ART2神经网络解决了传统ART2神经网络聚类中心漂移问题,杜绝了绝缘子故障漏判的发生.仿真实验结果表明,应用改进的ART2神经网络可有效实现绝缘子故障在线检测,获得较好的诊断结果.
According to the characteristics of insulator on-line detection, an improved Adaptive Resonance Theory (ART2) neural network was designed for the on-line diagnosis of faulty insulators. An initial layer was added in to the layer F2 of ART2 neural network, that is, the improved ART2 neural network divided layer F2 into upper-sublayer F22 and lower-sublayer F21. When insulators worked at the first time, upper-sublayer F22 saved the known patterns of the initial state and the typical-malfunction state. Simultaneously, lower-sublayer F21 stored the new typical-malfunction states. Firstly, the Corona Current Pulses of insulators were diagnosed with the corona current pulses on-line measurement instrumentation. Secondly, the N-ψ, map was drawn and was made the characteristic variable which was identified with the improved ART2 neural network. During the time, the characteristic variables were first compared with the known typical-malfunction state. If matching, they belonged to the sort; if not, they were compared with the new typical-malfunction states. A new malfunc- tion state would be made if they did not match all of them. The clustering center drifting problem of ART2 neural network were solved, and false diagnosis could be greatly reduced. Results of experiment have shown that this improved ART2 neural network has better performance and higher intelligence in the fault diagnosis of insulators.
出处
《湖南大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2008年第10期41-45,共5页
Journal of Hunan University:Natural Sciences
基金
国家自然科学基金资助项目(50377029)
山东省电力科技发展基金资助项目(2006A-22-3)
关键词
神经网络
ART2
绝缘子
在线检测
模式识别
故障诊断
neural network
ART2
insulator
on-line monitoring
pattern recognition
fault diagnosis