摘要
根据元件故障与保护动作和断路器跳闸之间的内在逻辑关系,建立了面向元件的电网故障诊断模型,并采用误差反向传播的梯度下降法修正网络参数。该模型是一种由 Noisy-Or,Noisy-And 节点组成的特殊的贝叶斯网络,能够处理电网故障诊断中的不确定性,具有语义精确、推理快速、学习效率高等特点,适用于大规模电力系统的多重复杂故障诊断。实际电网故障案例验证了该方法的正确性和有效性。
The proposed models are special Bayesian networks consisting of Noisy-OrandNoisy-Andnodes, which are converted from the logic relationship among section fault, protective relay operation and circuit breaker trip. The learning algorithm of the network parameters is analogous to the standard backpropagation algorithm used to train a multi-layered feedforward neural networks. The proposed approach can deal with uncertainties imposed on faultsectiondiagnosis in power systems.The models haveclear defined semantics,rapid reasoning,fast convergence, etc. Test results from a real sample power system have shown that the developed fault diagnosis models are correct and efficient, and are promising for on-line multiple fault diagnosis in large-scale power systems.
出处
《华北电力大学学报（自然科学版）》
CAS
北大核心
2004年第3期30-34,共5页
Journal of North China Electric Power University：Natural Science Edition
关键词
电网
故障诊断
贝叶斯网络
电力系统
专家系统
人工神经网络
fault section estimation
Noisy-Or node
Noisy-And node
Bayesian networks
parameter revision
