基于Gammatone滤波器倒谱系数与鲸鱼算法优化随机森林的干式变压器机械故障声音诊断

为有效提取变压器声音信号中的机械状态信息并识别其典型机械故障,依据人类听觉系统优异的声音识别能力,提出了一种基于Gammatone滤波器倒谱系数(GFCC)和鲸鱼算法优化随机森林(WA-RF)的变压器机械故障声音诊断方法。首先计算了变压器声音信号的GFCC,引入信息熵提取了GFCC中的主要声音特征信息。采用鲸鱼算法通过优化随机森林中决策树基分类器的规模和特征子集,构造了基于优化随机森林的变压器典型机械故障分类模型。对以某10 kV干式变压器正常与典型机械故障下声音信号的计算结果表明,所构建的基于GFCC主要特征参数和鲸鱼算法优化随机森林的变压器典型机械故障模型具有较好的识别效果,准确率可达95%以上,且具有优良的抗噪性能和鲁棒性。

Mechanical fault sound diagnosis based on GFCC and random forest optimized by whale algorithm for dry type transformer

To effectively extract the mechanical condition information of power transformer and then identify the mechanical faults via acoustic signals, a mechanical fault sound diagnosis method based on GFCC(Gammatone Filter Cepstral Coefficient) and random forest optimized by whale algorithm for transformer is proposed according to the excellent sound recognition ability of human auditory system. Firstly, the GFCCs of transformer acoustic signal are calculated, and the information entropy is introduced to extract the main acoustic feature information in GFCC. Then the whale algorithm is used to optimize the scale and feature subset of decision tree-based classifier in random forest and the classification model of typical mechanical fault based on optimized random forest is constructed. The calculative results of acoustic signals of a 10 kV dry-type transformer under normal condition and typical mechanical faults show that the typical mechanical fault model of transformer based on GFCC main characteristic parameters and random forest optimized by whale algorithm has better recognition ability with high accuracy of more than 95 %,and has excellent anti-noise performance and robustness.