基于LSTM神经网络的油浸式变压器异常声纹诊断方法研究

利用声音信号对电力变压器进行状态诊断是一种不停机、无接触的设备维护方法，可以诊断变压器异常状态类型。提出了一种基于LSTM神经网络的电力变压器异常诊断的方法，采集电力变压器在正常状态、过载和放电3种运行状态下发出的声音信号，将声音信号进行预处理并提取声音信号的MFCC特征，再将其通过一、二阶差分组合成一组声音特征的矢量，输入LSTM神经网络中进行训练。训练结果表明，将LSTM神经网络应用在电力变压器状态声音诊断上对3种状态的识别均能达到99%以上的准确率。

Abnormal Voiceprint Diagnosis Method of Oil-immersed Transformer Based on LSTM Neural Network

Using sound signals to implement the status diagnose of a power transformer is a method for non-stop and non-contact equipment maintenance， and can distinguish the abnormal status from the normal status of the transformer. The paper proposes a method of the transformer abnormal diagnosis based on a LSTM neural network. Firstly， the sound signal emitted by the transformer in the normal status and in overload and discharge conditions is collected，and the sound signal is preprocessed and the Mel-frequency cepstral coefficient（MFCC） features of the sound signal are extracted. Then the signals are grouped into a set of sound feature vectors through the first order and second order difference，and are input into the LSTM neural network for training. The training results show that the LSTM neural network can be applied to the acoustic diagnosis of the transformer status，and the recognition accuracy of the three status can reach more than 99%.