基于BP神经网络的电除尘火花放电识别

针对高压静电除尘中会发生火花放电现象，降低除尘效率、损坏设备的问题，提出了根据火花放电造成的声音进行火花识别的方法，利用MEMS（Micro-Electro-Mechanical System）数字麦克风采集声音信号，分析了火花放电声音的短时能量、短时过零率、线性预测倒谱系数和梅尔频率倒谱系数（MFCC）.建立BP神经网络识别系统，选用不同特征向量进行实验.研究结果表明：使用MFCC系数结合短时能量和短时过零率能提高识别率，对纯净样本的识别率高达96%，且用火花放电瞬间两帧数据作为火花样本进行BP神经网络训练能大幅度提高识别系统的鲁棒性，对非纯净样本的识别率高达95%.

Spark Discharge Identification of Electrostatic Dust Removal Based on the Back-propagation Neural Network

Spark discharge can be observed in high-voltage electrostatic dust removal, reducing the dust removal efficiency and damaging the equipment. In this study, we propose a spark identification method based on the sound because of spark discharge and use a microelectromechanical system microphone to obtainasound signal. Further, we establish a back-propagation (BP) neural network recognition system and conduct experiments using different characteristic vectorsafter analyzing the short-time energy, short-time zero-crossing rate, linear predictive cepstral coefficient, and melfrequency cepstral coefficient (MFCC) of sound. The experimental results denote that the application of MFCC based on short-time energy and short-time zero-crossing rate can improve the recognition rate, which is as high as 96% for pure samples, and that the application of instant spark discharge's two frame data as spark training samples for the BP neural network can considerably improve the robustness of the recognition systems; the recognition rate is observed to be as high as 95% in case of impure samples.