摘要
针对变压器运行状态声纹识别的应用需求以及BP神经网络识别模型准确率较低等问题,提出了一种改进人工神经网络的变压器声纹识别技术。该技术以变压器声纹信号中的MFCC系数作为模型的输入特征向量,在BOA算法中引入动态权重因子和变异因子对BP神经网络权值和阈值进行寻优,开展声纹识别。实验结果表明,利用变压器声纹信号的32维MFCC特征系数可使识别准确率达到90%以上,优化后算法的运算速度比PSO-BP神经网络与BOA-BP神经网络提高了9.24%和8.64%,具有更高的运算效率和识别准确率。
Aiming at the application demand of transformer voiceprint recognition and the low accuracy of BP neural network pattern recognition model,an improved artificial neural network technology for transformer voiceprint recognition was proposed.MFCC coefficient of the transformer voiceprint signal was taken as the input feature vector of the model,dynamic weight factor and variation factor were introduced into BOA algorithm,the weight and threshold of BP neural network were optimized,and voiceprint recognition was carried out.The experimental results show that the recognition accuracy can reach more than 90%by using the 32-dimensional MFCC characteristic coefficients of the transformer voiceprint signal.At the same time,the operation speed of the algorithm is 9.24%and 8.64%faster,respectively,than those of PSO-BP neural network and BOA-BP neural network,and it has higher efficiency of operation and recognition accuracy.
作者
李瑞琪
李燕
杜水婷
王军
LI Ruiqi;LI Yan;DU Shuiting;WANG Jun(School of Electrical and Electronic Engineering,North China Electric Power University,Baoding 071003,Hebei,China;Digitalization Division,State Grid Gansu Electric Power Company Co.,Ltd.,Lanzhou 730050,Gansu,China)
出处
《沈阳工业大学学报》
CAS
北大核心
2024年第4期380-387,共8页
Journal of Shenyang University of Technology
基金
甘肃省科技重大专项计划项目(19ZD2GA003)。
关键词
声纹识别
BP神经网络
特征向量
权重因子
动态寻优
模式识别
变异因子
状态检测
voiceprint recognition
BP neural network
feature vector
weight factor
dynamic optimization
pattern recognition
variation factor
state detection
