Research On Detection Technology Of Foreign Objects In Electric Energy Meter Based On Sound

As a national compulsory certification of electric energy measurement tool,it is of great significance to ensure the accuracy and reliability of electric energy meter.Traditional foreign object detection adopts the manual method of "hand-shake + ear-listen",which has low detection efficiency,low accuracy and potential safety hazards.In view of this,this thesis draws lessons from abnormal voice recognition technology,and collects the sound signal generated by the shaking of the watt-hour meter by using a sound card through a simulated artificial meter rocker device.Based on the analysis of the sound signal,a foreign object detection method in the watt-hour meter based on blind source separation and support vector machine model is proposed.The method can achieve a higher accuracy of foreign objects under the condition of complex sound components and small sample size,and it can realize automation of foreign objects detection.The foreign object sound signal in the watt-hour meter collected by sound card is often a mixed signal of multiple sounds.In order to improve the accuracy of foreign object detection,a blind source separation method based on single channel watt-hour meter for the separation of foreign body sound signals is presented.In view of the fact that the actual signal has some corre3 3lation and is not sparse enough,the method based on non-negative matrix decomposition was adopted and compared with the method based on independent component analysis.Firstly,the sound signal is divided into frames,then the spectrum amplitude matrix is obtained by Fourier transform,and finally the source signal is obtained by non-negative matrix decomposition.The simulation results show that the correlation coefficients between the source signals and the actual signals are 0.9674,0.9069 and 0.9584,respectively,when the signal has a certain correlation.This method has a better effect for separating foreign object sound signal in the actual electric energy meter.In order to obtain a higher recognition rate,it is necessary to detect the endpoint of the signal.In fact,the initial part of the collected sound signal is usually non-silent.In view of this,this thesis proposes an endpoint detection method based on fuzzy C-means clustering.Firstly,the Mel frequency cepstrum feature of sound signals are extracted.Secondly,the classification of silent and non-silent segments of sound signal is completed by using the method of fuzzy Cmeans clustering.The threshold-based endpoint detection method often leads to the degradation of endpoint detection performance due to the inaccurate threshold setting in the non-silent part of the initial part of the collected sound signal.The fuzzy C-means clustering method was used and compared with the traditional endpoint detection methods based onshort-time energy,short-time zero-crossing rate and improved spectral entropy.The experimental results show that the detection accuracy of the three methods based on shortterm energy and zero-crossing rate,spectral entropy and fuzzy C-means clustering is 0.9104,0.7889 and 0.7207,respectively.In addition,the objective function of the fuzzy C-means clustering method with short-time energy,short-time zero-crossing rate,spectral entropy and Mel frequency cepstrum coefficient were evaluated with different iteration times.The experimental results show that the convergence of the fuzzy C-means clustering method based on Mel frequency cepstrum coefficient has the least number of iterations,which can better describe the difference between silent and non-silent segments of sound signals.Finally,on the basis of investigating sound recognition technology,and combining with the fact that the actual sample size is relatively small,a foreign object detection system in the watthour meter was established based on SVM.The sound signal of watt-hour meter with foreign objects was investigated in detail and a large number of simulation experiments were carried out.According to the frame length and frame shift,sample number,characteristic parameters and model parameters,the best accuracy rate of foreign object detection was achieved with different parameters,such as frame length,sample number,characteristic parameters and model parameters,and the reasonable values of the above parameters were obtained.The accuracy of the foreign object detection is not less than 90%.