【摘要】 针对无人机巡检信号声纹特征提取过程受到多种外部环境因素影响的问题，提出一种基于小波包能量（WPE）-蝙蝠算法（BA）-卷积神经网络（CNN）(WPE-BA-CNN)优化模型的输电线路放电声纹特征识别方法。首先，利用小波包变换对采集到的声纹信号进行降噪处理，抑制低频噪声和传递路径干扰，提高信号质量并最大限度保留放电特征。然后，采用BA对CNN的卷积核数量超参数进行优化，增强模型的收敛速度和识别准确率。最后，通过实验验证所提方法的故障识别准确率。实验结果表明，与其它常用算法对比，所提方法在输电线路放电声纹识别具有较高的准确率和收敛速度。更多还原

【Abstract】 To address the issue of voiceprint feature extraction in UAV inspection signals being affected by various external environmental factors,the paper proposes a discharge voiceprint feature recognition method for transmission lines based on an optimized WPE-BA-CNN model. Firstly,wavelet packet transform is used to perform the noise reduction processing on the collected voiceprint signals,suppressing low-frequency noise and transmission path interference,thereby improving signal quality and preserving discharge features to the greatest extent. Then the bat algorithm(BA) is employed to optimize the hyperparameters of the number of convolutional kernels in the convolutional neural network(CNN),enhancing the convergence speed and recognition accuracy of the model. Finally,experiments are conducted to validate the fault recognition accuracy of the proposed method. Experimental results show that,compared with other commonly used algorithms,the proposed method achieves higher accuracy and faster convergence in the transmission line discharge voiceprint recognition.更多还原