RBF网络在小电流接地系统故障选线中的应用

The Application of RBF Network to Grounded Fault Line Selection in Non-effectively Grounded Neutral Mode

如何准确实现故障选线是小电流接地系统长期存在的难题,现有的选线方法和装置,都存在着许多不足。针对这种情况,以理想的建模为背景,以提高小电流接地系统的故障选线准确率为目的,提出了基于RBF神经网络的故障选线方法。RBF神经网络是一种局部逼近的神经网络,选取高斯基函数作为RBF基函数。文中在理想情况下建立一个模型,选取各条线路的零序电流、零序有功和零序无功作为输入的特征电气量,保证了故障线路特征选取的一般性;然后利用RBF神经网络强大的自适应、自学习能力,对电气特征量进行训练,保证了其快速的收敛性以及选线的准确性。文中的仿真结果表明,利用训练好的RBF神经网络可以实现故障选线,不但准确而且可靠,具有一定的可行性。

How to select the fault line quickly and accurately is a long-standing issue in non-effectively grounded neutral mode. There are many weaknesses in the selection methods and equipments nowadays. In allusion to this situation, a method of fault line detection based RBF neural network is put forward in this paper according to the background of ideal model and the aim of improving the rate of fault line detection in non-effectively grounded neutral mode. The RBF neural network is a kind of local approach neural net work, and its basic function is the Gaussian basic function. A model is constructed ideally. The zero sequence current, the zero sequence active power and the zero sequence reactive power are chosen as import electric-eigenvectors of every line; so the generality of the fault line can be guaranteed. Then the electric-eigenvectors are trained by the powerful ability of auto-adaption and self-learning. The simulated results indicate that it is accurate and credible to achieve grounded fault line detection by the trained RBF neural net work. so this method is feasible.