基于纵横交叉优化BP神经网络的变压器故障诊断研究

Study on Power Transformer Fault Diagnosis Based on Back-propagation Neural Network Optimized by Crisscross Optimization Algorithm

为提高电力变压器故障诊断的准确率,提出一种基于纵横交叉算法改进BP神经网络的故障诊断方法。该方法在BP神经网络结构的基础上,利用纵横交叉CSO算法对BP神经网络的权值和阈值进行优化,得到最优的权值和阚值向量,并将优化值代入BP神经网络训练模型中,然后利用BP神经网络的自学习功能进行训练,最终得到基于CSO-BPNN的变压器故障诊断模型。将提出的基于CSO-BPNN算法的故障诊断结果与标准BP神经网络算法故障诊断结果进行对比。测试结果表明,CSO-BPNN算法融合了CSO算法和BPNN算法的优点,能更有效地提高变压器故障诊断的识别精度,具有良好的工程实用价值。

In order to improve the accuracy of power transformer fault diagnosis,an improved back-propagation neural network optimized by crisscross optimization algorithm(CSO-BPNN) diagnostic approach is proposed in this paper.The CSO-BPNN algorithm is established based on the BP neural network structure.Meanwhile,the CSO algorithm is introduced to optimize the weights and threshold values of BP network.In CSO-BPNN,when the optimal weights and threshold vectors are obtained,the optimal values will be sent to the BP neural network.Then the training mechanism of BP neural network algorithm is executed to obtain the CSOBPNN based transformer fault diagnosis model.At the end,the proposed CSO-BPNN diagnostic approach is compared to the standard BPNN diagnosis approach,and the results show that the proposed CSO-BPNN approach integrates both the advantages of CSO algorithm and BPNN algorithm,can greatly improve the accuracy of transformer fault diagnosis,and has high value in engineering practice.