Research On Feature Extraction And Automatic Classification Of Mine Microseismic And Blasting Vibration Signals

The microseismic signals of coal and rock contain a large amount of useful information about coal and rock mass damage.Recognizing microseismic signals can locate the location of coal and rock fractures,which is of great significance for early warning of coal mine dynamic disasters.As the current coal mining environment is relatively harsh,the microseismic signals received by the sensors are often mixed with a large number of blast vibration signals,which greatly increases the difficulty of identifying microseismic signals.This paper focuses on the automatic classification of mine microseismic and blast vibration signals,starting from the time domain,frequency domain,and time-frequency domain,then the difference characteristics of the two types of signals are analyzed and extracted.Finally,an automatic identification model based on Particle Swarm Optimization-Support Vector Machine(PSO-SVM)is constructed to realize automatic classification and identification of microseismic and blast signals.The main research results and conclusions are as follows:(1)Aiming at the "under-segmentation" problem of traditional empirical wavelet transform(EWT)in processing complex signals spectrum,a new improved method on the basis of traditional EWT is proposed in this paper and then build a simulation signal for experimental analysis to verify the effectiveness of the improved algorithm.The results show that the effect of improved EWT on signal decomposition is better than that of traditional EWT and empirical mode decomposition(EMD)algorithms.(2)STA/LTA and Fast Fourier Transform(FFT)are used to analyze and count the duration and main frequency characteristics of microseismic and blast signals.The results show that the duration of the microseismic signals is mainly distributed within 2.2～4.0s,the microseismic signals contain more energy and its decay rate is slower,while the duration of the blast signals is generally maintained within 0.4～2.5s.The main frequency of microseismic signals is comparatively low which mainly around 20～30Hz,while the frequency spectrum of the blast signal is more complicated and its main frequency is higher than that of the microseismic signals which mainly in the range of 110～130Hz.(3)The improved EWT is used to decompose the microseismic and blast signals,then using the correlation analysis to screened out the optimal modal components and the timefrequency domain characteristics of the signals are extracted with the multiscale permutation entropy(MPE).The results show that there is a big difference between the MPE values of the microseismic signals and the blast signals,and the MPE values of the blast signals are larger than that of the microseismic signals.(4)The automatic recognition model based on PSO-SVM is constructed,and the multidomain fusion feature database of mine microseismic signals are used as the input of the PSOSVM model,then comparing and analyzing the classification accuracy of SVM for the two types of signals under different types of feature quantities.The results show that the timefrequency domain feature vector extracted based on the improved EWT＿MPE is used as the input of the PSO-SVM classification model and its classification and recognition accuracy reached 93.0%,which the accuracy respectively is 2% and 4.0% higher than the traditional EWT＿MPE and EMD＿MPE,it means that the improved EWT algorithm proposed in this paper is feasible.And based on the improved EWT and combining the multi-domain fusion features of the two types of signals as the input of PSO-SVM,the recognition accuracy rate reached94.5%,which is 1.5% higher than the recognition accuracy rate that relies on a single timefrequency domain feature vector,it means that there is still limitations in the classification and recognition of microseismic and blast signals using a single feature vector.