| Microseismic monitoring technology uses microseismic sensors to capture the signals released by rock cracks,and obtains source parameters such as earthquake onset time,source location,event intensity,etc.by analyzing the collected microseismic signals,so as to provide early warning of possible dangers;As a passive monitoring technology,about 98%of the collected data are interference noise signals,and the monitoring environment is often harsh,resulting in a low signal-to-noise ratio of the acquired data.At present,the commonly used automation methods are greatly affected by the signal-to-noise ratio.When the signal-to-noise ratio is low,it is easy to cause misidentification and mispicking,and it is difficult to meet the high-precision engineering requirements.Based on this,the automatic identification method and automatic first arrival picking method of high-precision,high-efficiency and not easily affected by the signal-to-noise ratio of the micro-seismic event are particularly urgent;In addition,since the signal-to-noise ratio of the original microseismic monitoring data is relatively low,improving the signal-to-noise ratio is very important for the subsequent steps such as picking and positioning of the first arrival.Focusing on the above-mentioned problems,this article mainly studies the three aspects of microseismic signal identification,noise suppression,and pick-up of the microseismic signal P-wave arrival:Aiming at the problem of automatic identification of microseismic signals,this paper studies and compares the characteristics of all kinds of waveforms,and completes a convolution neural network model to realize the classification of monitoring signals by means of data enhancement and cross validation in the case of limited training data set.The trained CNN model can learn the main features of microseismic signals,blasting signals,and mechanical noise signals,and classify them through the differences in features.Finally,the accuracy of CNN in the test set can reach 98.7%,and the loss value can reach 0.09.When compared with the recognition results of STA/LTA and AIC,the accuracy and recall rates of CNN classification are 96.6% and 95.8%,respectively,which are better than the other two algorithms.In the identification of real monitoring data,the accuracy and recall rate of the microseismic signal constructed in this paper are 90.8% and 96.5%,which can better meet the requirements of engineering identification accuracy.Aiming at various noise signals generated in the process of signal acquisition,a new threshold function is proposed based on the study of wavelet denoising principle and four different threshold functions.Through the denoising simulation experiments of the noisy Doppler signal and the simulated microseismic signal,the denoising effects of several threshold functions are tested and analyzed numerically.The effectiveness of the improved threshold function is verified by comparing the denoising effects of waveform,SNR and RMSE.Finally,by comparing the performance of five denoising functions on real microseismic signals,it proves the anti-noise and accuracy of the improved threshold function.In order to solve the problem of automatically picking up the first arrivals of microseisms with low signal-to-noise ratio,this paper implements a picking algorithm based on long and short-term memory networks.Through the picking comparison experiment on simulated microseismic data,when the signal-to-noise ratio is low,the picking error of the algorithm in this paper is smaller than STA/LTA,M-AIC.For the real micro seismic data picking,the average error value of P-wave picking is 4.62 ms,of which the number of P-wave pick-up errors in the range ±3ms accounts for 24% of the total,62% of the total in the range±5ms,83% in the range ± 10 ms and 100% in the range ± 15 ms,which basically meets the engineering requirements.It can be seen that the cyclic neural network has certain effectiveness and accuracy in the pick-up of P-wave time of micro seismic signals. |
PDF Full Text Request