摘要
为根据局部放电信号识别早期的GIS绝缘故障缺陷类型,提出了一种利用ART神经网络在线识别GIS绝缘故障类型的新方法。较之常用BP神经网络,该法训练时间短、所需样本少、权值稳定、不存在局部收敛,故更适于在线识别。网络的输入量为一个工频周期内局部放电脉冲重复率、主频率、阻尼系数、放电量、放电相位分布。利用5种GIS绝缘缺陷类型的实验所得数据对ART神经网络进行训练及验证,证明该法的缺陷类型正确识别率可>98%,在GIS绝缘故障类型的在线模式识别中具有广泛的前景。
A new on-line pattern recognition method for GIS insulation defect is proposed based on the Adaptive Resonance Theory Neural Network （ARTNN）.Compared with Back Propagation（BP） NN,the proposed method has some merits of better performance of recognition such as shorter learning time,less number of training samples,stabilization of weight,and no local extremum.The input vector consists of the Partial Discharge（PD） pulses characters,e.g.the numbers of PD,repetitive rate of PD,quantity of PD,phase of PD,characteristic frequency of PD and damping coefficient of PD within one power frequency cycle.The scheme has been verified by utilizing experimental data derived from five sorts of GIS insulation defects,and the correct ratio could be up to 98%.Therefore,the proposed method has an abroad application in on-line recognition of GIS insulation defect.
出处
《高电压技术》
EI
CAS
CSCD
北大核心
2007年第12期75-79,共5页
High Voltage Engineering
