一种基于常规保护原理的ANN发电机差动保护研究

文中提出了一种基于常规保护原理的神经网络差动保护方法。神经网络具有非常优良的模式识别能力。文章首先从理论上分析得出，可以由一个单神经元实现差动保护中常规的比率制动特性，并实现了由一个单神经元感知器构成的比率制动特性的差动保护，在此基础上进一步提出了一个具有非线性制作特性的多层神经网络差动保护模型。该方法将传统保护中的整定值和人工神经网络结构中的权系数对应起来，利用学习算法获得保护的最佳方案，运行人员可以根据经验来选择训练结果，它彻底解决了将ANN技术应用到继电保护工程实际中去后可靠性差的问题，且容易被现场运行人员所接受。该方法无论在工程中还是在理论上都具有重要的意义。

A NEW METHOD OF ANN FOR DIFFERENTIAL PROTECTION BASED ON CONVENTIONAL PROTECTION THEORY

In order to optimize the performance of digital protection for generator transformer unit,a new method of ANN differential relay based on conventional protection theory was developed.ANN has a good capacity of pattern recognition.It was concluded from theoretic analysis that a perception neuron could realize conventional differential protection based on ratio restrained theory.The neuron model was simulated according to this theory.Furthermore a multi layers ANN of differential protection model was advanced which was characterized with non liner restrained theory.In this model,relay settings were corresponding to the weights of ANN.The problem of protection reliability in practising ANN technique could be solved completely by the optimal protection scheme achieved by learning rules.It was easy for site operators to accept the method and could be of value in theory as well as in practice.