Research On Microseismic Event Recognition Technology Based On Convolutional Neural Network

Microseismic monitoring technology consists of data collection,processing,event detection,analysis and evaluation,among which microseismic event detection is a very critical basic work.Microseismic detection has the characteristics of a long-term and complex collection environment,which causes problems such as low signal-to-noise ratio(SNR),large data volume,and an effective signal that is not obvious.Traditional event detection methods face significant challenges due to these issues.Recently,deep learning techniques have achieved some progress in the research of microseismic data processing and recognition.Professionals and scientists at domestic and international research institutes are concerned about how to use deep learning to detect microseismic events efficiently and accurately under the condition of strong noise interference and weak effective signal.In this paper,the above problems have been investigated.First,from the perspective of time series feature extraction,we can improve the extraction effect of microseismic time series features by building a multiscale neural network based microseismic event recognition model.Second,from the perspective of multimodal feature learning,a multimodal neural network microseismic event recognition model is constructed by combining the temporal features and S-domain features.The multidimensional modal features of microseismic signals are used to recognize effective events.The main contents are as follows:A method of microseismic event detection based on a multiscale detection neural network is proposed to solve the dilemma that the effective signal is easily disturbed by noise,resulting in a drop in detection accuracy.Initially,the one-dimensional convolution neural network is built according to the characteristics of the microseismic signal to extract the fine-grained features of the shallow layer and the semantic features of the deep layer of the microseismic signal.Then the credibility model is established for the detection results of feature expressions in different scales,and the results of final recognition are obtained through uncertainty reasoning.In comparison to the wavelet analysis method,BP network,and convolution neural network,the experimental results demonstrated the model is superior to other ways and has better anti-noise performance.A multimodal neural network based microseismic event detection method is proposed to address the problem that the time series characteristics of effective microseismic signals have severe limitations.First,the multichannel time-domain mode with the target channel as the axis symmetry is established by correlation of the collected data,and the S-domain modal characteristics are obtained by time-frequency analysis of the target channel.Then,the neural network for microseismic event detection is designed by combining the time-domain mode and S-domain mode.It synthesizes multimodal features for training and learning to improve the accuracy of microseismic event recognition.Finally,low SNR data analysis,small amplitude data analysis and actual oil well microseismic monitoring signal event analysis are performed on the synthetic microseismic signal to verify the effectiveness of the method.It can be concluded that the method in this paper has higher noise resistance and accuracy,through comparative experiments with SVM,CNN,supervised machine learning algorithm and the method in this paper.In this paper,the problems of the current methods of detecting microseismic events based on deep learning are analysed.Design of a multiscale detection model and a multimodal detection model.They can effectively improve the detection accuracy of effective events of microseismic signals and lay the foundation for subsequent microseismic monitoring work.