基于GA-BP混合算法的电力变压器故障诊断方法

为解决BP神经网络在变压器故障诊断中存在的收敛速度慢,容易陷入局部最优点等缺点,采用了将BP网络和遗传算法相结合的方式,利用遗传算法的全局收敛性,优化BP网络的初始权值和阈值,再由BP网络进行调整搜索,同时采用了LM优化方法训练神经网络以提高网络精度,缩短训练时间,最后将训练好的网络应用到油中溶解气体分析技术中。

Method on Power Transformer Fault Diagnosis Based on GA-BP Hybrid Algorithms

In order to solve own weakness of BP neural network,such as slow convergence rate and often converges to local optimum,the method that BP network combined with genetic algorithm was adopted,the initial weight value and threshold value of BP neural network was optimized by using global convergence of genetic algorithm,then was adjusted and searched by BP network,after that,the LM algorithms was introduced to training BP neural network for improving precision of network and shortening training time,finally,the trained network was applied to dissolved gas analysis technology.