基于径向基函数神经网络和模糊控制系统的电网故障诊断新方法

HYBRID FAULT SECTION ESTIMATION SYSTEM WITH RADIAL BASIS FUNCTION NEURAL NETWORK AND FUZZY SYSTEM

该文针对RBF神经网络的知识存储和诊断过程是一个黑箱,对运行人员不透明,且当电网拓扑结构发生变化或扩展时,神经网络只能重新训练等问题,推导并建立了RBF神经网络和模糊控制系统之间的等值关系,使得蕴含在RBF神经网络权重中的知识转变为等值模糊控制系统中用语言表述的规则。在此基础上,针对电网结构发生变化或扩展情况,提出了RBF神经网络的局部重新训练新算法。提出的基于RBF神经网络和等值模糊控制系统的故障诊断方法在IEEE118母线系统中进行了仿真试验,结果表明:基于RBF网络与等值模糊系统的故障诊断方法诊断知识易于理解,诊断过程透明,并能适应电网拓扑结构发生变化或扩展的情况,效果理想。

In this paper, functional equivalence between a radial basis function neural networks （RBF NN） and a companion fuzzy system （CFS） is buik up throughout the neural network training process, therefore the black-box-like knowledge in a RBF NN will be rule-based and transparent in its CFS. Through useful knowledge extraction from the old CFS and insertion back to the new CFS piece by piece, the RBF NN retraining issue under network expansion and topology change can be solved effectively and efficiently. The corresponding FSE system has been implemented and tested in the IEEE 118-bus power system. The simulation results show that the suggested approach for RBF NN retraining works successfully and efficiently in the case of power network expansion and topology change, which significantly improves the application potential of RBF NN in FSE of practical power systems.