Research On Microseismic Event Detection Method Based On Convolutional Neural Network

With the rapid development of global technology and economy,the development of unconventional oil and gas reservoirs,such as tight oil and gas,shale oil and gas,and coalbed methane,has become more and more important.In the field of unconventional oil and gas exploration and development,hydraulic fracturing technology is a necessary mean to improve oil and gas recovery,and microseismic monitoring is a key technology to evaluate the effect of fracturing.Microseismic monitoring technology refers to the processing of seismic waves released by formation fractures during the fracturing process,interpreting information such as the fracturing area and fracture development direction,and then objectively evaluating the fracturing effect.During hydraulic fracturing construction,microseismic monitoring technology requires real-time observation of fracturing fractures so that engineers can adjust the mining plan in time based on fracturing feedback.This needs timely and effective processing of microseismic monitoring data.In the process of monitoring data processing,the purpose of microseismic event detection is to eliminate noise interference signals and identify effective microseismic waveforms.Therefore,the rapid and accurate detection of microseismic events is the basis for subsequent data processing.This thesis aims to propose a new method that can automatically and efficiently detect microseismic events.The main work can be divided into the following parts:First,the principles of downhole microseismic monitoring technology and surface microseismic monitoring technology are studied,and then the advantages and disadvantages of both are briefly analyzed.Based on the surface microseismic monitoring data,the source and waveform characteristics of effective fracturing signals and various random interference signals are analyzed to design a reasonable ground microseismic monitoring data processing method.The applicability of the method is analyzed by using the long-short time window ratio(STA/LTA)method to detect high and low signal-to-noise ratio ground microseismic actual monitoring data,and the influence of the selection of trigger threshold and other parameters is discussed on the detection effect.The advantages and disadvantages of this method are summarized,which provides a basis for the subsequent proposal of new microseismic event detection methods.Second,a microseismic event detection method is proposed based on convolutional neural network(CNN)to solve the problems of cumbersome preprocessing steps and serious manual intervention in traditional microseismic event detection.For the training and testing of a CNN,a sample data set is constructed from actual multi-site microseismic monitoring data of an oil well’s hydraulic fracturing,which consists of valid event signals and invalid background noises and their classfications.Then,the CNN is trained and tested by utilizing the sample data set,and an optimal CNN model is obtained with best accuracy of event detection.To test the performance of the CNN model,synthesized microseismic signals with different signal-to-noise ratios,and actual monitored microseismic signals of several oil and gas wells’ hydraulic stimulation are fed into the CNN model for event detection.The processing results demonstrate that the CNN model can automatically detect microseismic events effectively,and has good abilities in noise suppressing and generalization.Third,a microseismic event detection method is proposed based on time-frequency analysis and deep convolutional neural network(CNN),due to the fact that the time-frequency analysis method can effectively reflect the advantages of the integrated information of the microseismic signal in the time-frequency domain.In the method,S transform is first used to extract the time-frequency spectrum of hydraulic fracturing microseismic monitoring signals of oil and gas wells,and a sample data setis constructed,then a deep convolutional neural network is constructed to realize the feature extraction and classification of samples.Compared with the algorithms combining CNN with other time-frequency analysis methods such as short-time Fourier transform and wavelet transform,the detection method based on S-transform and CNN has higher recognition accuracy and stability.In order to verify the feasibility of the proposed method,the synthetic microseismic signals with low signal-to-noise ratio(SNR),and different types of surface microseismic monitoring signals of oil wells are detected,respectively.The processing results show that the novel method can effectively detect multiple types of microseismic events,including low SNR signals and weak signals.