Gas Tank Air Leakage Identification in Power Plant Based on Subdivided Self-encoding Features

The equipment of hydropower plant is widely distributed underground, and it is difficult to achieve full-time and full-cycle coverage only by manual inspection. With the continuous development of intelligent and intelligent technology, unattended power plants have become possible. The pressurized gas tank in the production area of the power plant has a relatively complex distribution of working conditions: normal operation, normal deflation, and normal shutdown. Due to the similarity in the sound scale between the leakage fault of the pressurized gas tank and the normal gas leakage, this paper proposes a long short-term memory-convolution sound recognition network based on stacked autoencoder features. First, the distribution characteristics of the sound of the pressurized gas tank are analyzed. Secondly, the pre-processed voice is subdivided and encoded by the stack autoencoder. Finally, a long short-term memory(LSTM)-convolution parallel network is constructed to identify the air leakage sound. The test results show the high accuracy and effectiveness of the method.