基于小波散射网络-贝叶斯优化门控循环单元的电力变压器声纹识别方法OA

This paper proposes a voiceprint recognition method for power transformers with small-scale samples based on wavelet scattering network-Bayesian optimized gated recurrent unit(GRU).Firstly,in order to filter out interference components and improve the accuracy of voiceprint recognition,the original signal extracted from the transformer is subjected to blind source separation through empirical wavelet transform(EWT)and fast independent component analysis algorithm(FastICA),resulting in the voiceprint signal of the transformer itself.Then,the feature vector of the voiceprint signal is extracted using the wavelet scattering network as the input of the voiceprint recognition model,and a GRU is applied as the classifier.To improve the recognition accuracy,Bayesian algorithm is utilized to optimize the hyperparameters of GRU layers and initial learning rate.The experimental results show that in the case of small sample size,compared with the commonly used voiceprint time-frequency spectrum and deep convolutional neural network,the model constructed in this paper converges faster and the recognition accuracy increases,significantly improving its performance.