The sound signal generated by GIS mechanical fault contains a large amount of equipment operation state information. The signal analysis method based on voiceprint recognition technology is an effective means to realize GIS live detection and fault diagnosis. In this paper

LPCC

and

MFCC

feature vectors under different operating conditions of GIS are extracted. The calculation results of

F

ratio show that

MFCC

is a more differentiated feature vector. On this basis

using the global search ability of particle swarm optimization algorithm and the local adjustment ability of evolutionary programming algorithm

focusing on particle swarm optimization and introducing the mutation operation of evolutionary programming algorithm

a hybrid optimization method based on vector space state optimization is formed. The vector quantization model proposed in this paper is used to detect and recognize the sound signals under different operating conditions of 110 kV GIS. The experimental results show that the codebook distortion obtained by this scheme has smaller mean and variance

better optimization performance and stability

and the recognition accuracy under the same conditions is about 90% ~ 96%

which is better than the classical LBG iterative algorithm.