Intelligent Recognition And Location Of Geohazard Microseismic Via Deep Learning

Geohazards have always posed a threat to people’s lives and property due to their irregular occurrence.People urgently need to monitor and warn geohazards scientifically and effectively.As one of the warning signals for geohazards,microseismic can be recognized and located through microseismic monitoring technology,thus realizing the prediction of the possibility of geohazards.The main steps of microseismic monitoring technology is microseismic signal recognition,phase association and hypocenter location.For the traditional methods used in microseismic monitoring technology which are simple and the calculation efficiency are high,they are too dependent on parameter settings and have poor flexibility in the processing,resulting in unsatisfactory results.Therefore,this thesis proposes the use of deep learning methods to compensate for the shortcomings of traditional methods,while achieving more intelligent and accurate microseismic signal recognition and source positioning than existing deep learning methods,providing a more convenient and reliable solution for effective monitoring and early warning of geological disasters in the future.It has practical significance for the people in geological disaster prone areas to reduce disaster reduction and economic losses.The specific researches are as follows:(1)Microseismic signal recognition: The research on microseismic signal recognition based on deep learning has gradually matured,but the labels used to help neural network training and learning still rely heavily on manual labeling,which is cumbersome and has poor self-adaptability.Therefore,this thesis proposes microseismic intelligent recognition method,FC-Net,based on Fuzzy C-means clustering algorithm(FCM).The FCM used to adaptively label the microseismic phase arrivals,and the attention gates and recurrent residual convolutional units are added to the U-Net foundation to upgrade the network model,making it more accurate and intelligent for microseismic signal recognition.In practical applications,the proposed method has shown excellent results,proving that it not only has strong robustness to microseismic signals with low signal-to-noise ratio,but also improves its universality.(2)Microseismic hypocenter location: Traditional microseismic hypocenter location methods are dependent on the choice of velocity model and initial conditions,while deep learning can reduce the interference of human factors in location tasks.However,the lack of microseismic catalogs currently limits the development of deep learning in the field of microseismics.Based on this,this thesis proposes a method,Loc Net,based on transfer learning and multi-channel fusion model.By using multi-channel fusion model,this method can extract microseismic waveform features from each channel while also fusing and sharing parameters with each other.It also uses transfer learning to transfer the pretrained model weights on the seismic data to a new model,retrain on the microseismic data,and fine tune the model weights.The results of practical application confirm that the method has higher positioning accuracy.Deep learning can not only process microseismic signals more flexibly and intelligently,but also improve the stability and accuracy.In the future,it will still be an important development direction in the field of microseismic and even in the whole field of seismology and geology.