基于GA与L-M优化算法的变压器故障诊断研究

Transformer fault diagnosis based on GA and L-M optimization algorithm

利用MATLAB环境建立一个用于变压器故障诊断的BP网络模型。首先利用具有全局寻优功能的遗传算法对BP神经网络的初始权值和阈值进行优化,然后采用L-M(Levenberg-Marquardt)优化算法对BP神经网络进行训练,从而达到加快网络训练速度,避免训练过程陷入局部极小点的目的。最后,详细记录网络的实际输出,并与期望输出做对比研究,最终证实了此网络达到了设计要求,可用于变压器的故障诊断。

A BP network model for transformer fault diagnosis is established based on the MATLAB environment. Firstly, the initial weights and thresholds of BP neural network are optimized by genetic algorithm which has the feature of global optimization; Then, L-M （Levenberg-Marquardt） algorithm is used for training BP neural network,so the network training speed is increased and local minimum points is avoided in the training process. Finally, the actual output is gained and made comparative study with the expected output. Finally, it confirms that this network model has a high accuracy and can be used for transformer fault diagnosis.