一种早期诊断变压器故障的神经网络方法

Using BP neural network to diagnose transformer's forepart troubles

根据罗杰斯三比值法的理论 ,建立了变压器故障诊断的BP网络模型 .采用启发式学习算法 ,减少了学习过程的振荡趋势 ,改善了收敛条件 ;采用改进的自适应调整学习率方法 ,减少了出现局部极值的概率 .在MatLab环境下 ,应用该网络模型 ,对某台大型电力变压器进行早期故障诊断 ,误差为 0 .0 0 2 0 ;对某台小型调容变压器进行早期故障诊断 ,误差为 0 .0 0 19.结果表明 :该模型可以指导实际生产 。

In order to meet the demand of forepart diagnosis transfermer's trouble. By using BP neural network method, this paper introduces Royce three-ration method that is getting much attention in analytical method of gas-behind-oil, then the transfermer's trouble diagnosis model (ANN1) is found. On MatLab condition, we train the BP network to diagnosis 2 example transfermer's trouble according to three-ration method. The result shows that it accords with practice.