基于特征降维和神经网络的电能表内异物声音自动识别

电能表是国家强制检定的电能计量工具,其计量的精确性影响着千家万户的利益。传统的人工检测方式不仅效率低而且检测结果不稳定。随着声学检测技术的日趋成熟,采用声学检测的方式来检测电能表内的异物已成为实现工厂自动化的大势所趋。针对现有半自动的人工检测电能表异物方式,提出一种基于特征降维和神经网络的电能表内的异物声音自动识别方法。该方法充分利用声音的时、频域特征系数和倒谱系数,先对声音信号进行通道转换、预处理和数字降噪,再对声音信号进行时、频域和倒谱分析,并同时提取其短时特征系数及改进后MFCC系数。将声音特征通过PCA降维后输入基于Adaboost算法聚类后BP神经网络分类识别,并与传统的BP神经网络分类进行比较,证明了该方法的有效性。这里给出了电能表异物自动识别技术实现的具体步骤,并通过MATLAB仿真实验证明了该方法的有效性,BP神经网络的平均识别率较高,可达到95%以上,并且计算复杂度小易于实现。

Automatic Recognition of Foreign Object Sound in the Electricity Meters Based on Feature Dimension Reduction and Neural Network

The electricity meter is a kind of power metering tool subject to compulsory verification,whose accuracy of measurement affects the interests of users.The traditional manual detection method is not only inefficient but also unstable.With the development of acoustic detection technology,it is a widespread trend to use acoustic detection to detect the foreign objects in the electricity meter.In this paper,an automatic recognition method of foreign object sound in the electricity meters based on feature dimension reduction and neural network is proposed to improve the detection of foreign objects in the electricity meters.This method makes full use of time and frequency domain characteristic coefficients and cepstral coefficient of sound,which firstly performs channel switching,pre-processing and digital noise reduction for audio signals,and then carries out time,frequency domain and spectrum analyses for audio signals,and extracts its short-time characteristic coefficient and modified MFCC coefficient.The sound features are input into BP neural network based on Adaboost algorithm clustering after PCA dimension reduction for classification recognition.The effectiveness of this method is demonstrated by comparing with conventional BP neural network classifications.The specific steps to realize the automatic recognition technology of foreign object in the electricity meters are given,and the effectiveness of this method is proved by MATLAB simulation experiment.The average recognition rate of BP neural network is higher than 95%,and the computational complexity is low and easy to realize.