动态对角递归网络在变压器故障诊断中的应用

Application of diagonal recurrent neural network in power transformer fault diagnosis

介绍了动态对角递归网络,并针对BP算法收敛慢的缺点,提出了递推预报误差学习算法。利用该算法对神经网络的权值和域值进行训练,有效地提高神经网络的收敛性及增量学习能力。将动态对角递归网络应用到变压器的故障诊断中,利用改良三比值方法来实现诊断,建立了诊断的模型。利用部分数据进行了训练及故障诊断的仿真,结果表明了利用该方法进行变压器故障诊断的有效性。

A simple recurrent neural network named diagonal recurrent neural network is studied in this paper. To overcome the slow convergence of the BP algorithm, recursive prediction error algorithm is proposed, which can train both the weight and the bias. This algorithm can improve the astringency and increment learning ability of the neural network effectively.In addition, the recurrent neural network trained with RPE algorithm is used in fault diagnosis of power transformer based on improved three-ratio method.A model of fault diagnosis based on DRNN is established. Some data are used to simulate, the simulation diagnosis demonstrates the effectiveness of the proposed algorithm.