基于长短时记忆神经网络在故障故障中参数依靠经验确定，具有不确定性，从而

基于长短时记忆神经网络在故障故障中参数依靠经验确定，具有不确定性，从而导致模型的故障诊断准确率降低等缺陷，本文采用改进的灰狼算法对长短记忆神经网络进行优化，提出一种基于改进灰狼算法的LSTM网络参数优化OLTC故障诊断方法。首先采用多尺度加权排列熵和基于GOA-VMD分接得到的能量熵作为LSM的输入；然后通过使用改进的灰狼算法对LSTM的相关超算数进行迭代优化；最后构建IGWO-LSTM组合模型对OLTC中的不同故障进行分类。该模型克服了依据经验选取参数而导致分类精度角度的问题。算例分析表明：相对于传统分类算法，所提方法更好的对OLTC的不同故障进行分类，提高了分类精度，为OLTC在线监测进一步提供了理论依据。

基于长短时记忆神经网络在故障故障中参数依靠经验确定，具有不确定性，从而导致模型的故障诊断准确率降低等缺陷，本文采用改进的灰狼算法对长短记忆神经网络进行优化，提出一种基于改进灰狼算法的LSTM网络参数优化OLTC故障诊断方法。 首先采用多尺度加权排列熵和基于GOA-VMD分接得到的能量熵作为LSM的输入；然后通过使用改进的灰狼算法对LSTM的相关超算数进行迭代优化；最后构建IGWO-LSTM组合模型对OLTC中的不同故障进行分类。 该模型克服了依据经验选取参数而导致分类精度角度的问题。 算例分析表明：相对于传统分类算法，所提方法更好的对OLTC的不同故障进行分类，提高了分类精度，为OLTC在线监测进一步提供了理论依据。

Based on the fact that the parameters of long-term and short-term memory neural network are determined by experience and uncertain, which leads to the decrease of fault diagnosis accuracy of the model, an improved grey Wolf algorithm is used to optimize the long-term and short-term memory neural network, and an OLTC fault diagnosis method based on Improved Grey Wolf algorithm is proposed. Firstly, the multi-scale weighted permutation entropy and the energy entropy based on goa-vmd demultiplexing are used as the input of LSM; secondly, the improved gray wolf algorithm is used to iteratively optimize the relevant super arithmetic of LSTM; finally, the igwo-lstm combination model is constructed to classify different faults in OLTC. The model overcomes the problem of selecting parameters based on experience, which leads to classification accuracy angle. Example analysis shows that: compared with the traditional classification algorithm, the proposed method can better classify different OLTC faults, improve the classification accuracy, and provide further theoretical basis for OLTC online monitoring.<br>