Intelligent Microseismic Location Method

Microseismic monitoring is an important technical method in geophysical exploration,such as coal mining and natural disaster assessment.Microseismic location is the key part of microseismic monitoring.In the past,most of the commonly used methods for microseismic location are traveltime-based methods.It constructs a mathematical equation based on the difference between the actual travel time and the theoretical travel time,and solves the function value to get the source location.This method needs to obtain the initial time from microseismic events to different geophones,which involves the operation of microseismic signal recognition and first arrival extraction.In these processing processes,the traditional methods often need to set parameters manually or obtain additional prior knowledge.And traditional method in the face of low signal-to-noise ratio of micro seismic events is difficult to have good performance.Therefore,an intelligent microseismic location method is implemented in this thesis to avoid tedious parameter setting.Convolutional neural network can be used to extract waveform features automatically.In order to solve the problems of low signal-to-noise ratio of microseismic events and lack of effective microseismic event catalogue,corresponding improvements are proposed,which are mainly divided into signal recognition and event location:(1)Microseismic signal recognition: The introduction of the convolutional neural network makes microseismic signal classification become intelligent,without setting thresholds and human participation,etc.But normal network classifier is as difficult as traditional method to deal with low SNR scenes.Therefore,this thesis implements a dual-channel convolutional neural network combining time-domain signal and wavelet domain information.The wavelet packet decomposition coefficient provides the information of microseismic signals in time-frequency domain for network classifier,which can be used as the auxiliary information of time-domain signals for classification learning.Finally,the model shows high recognition accuracy in low SNR microseismic data sets and field data sets.It is proved that the model has anti-noise property and can be used in actual data processing.And the performance of the model is stable.(2)Microseismic events location: the traditional location method needs to obtain the velocity model and other parameters of the survey area,which adds a lot of extra work to the location research.In the field of seismology,some scholars use neural network to locate earthquake sources.Since the seismic network is very mature and the earthquake event catalogue can be easily obtained,most of the related researches are based on supervised learning.However,the situation in microseismic field is quite different.The lack of an effective event catalog makes it difficult to implement supervised learning methods.Therefore,this thesis implements an intelligent location method based on semi-supervised learning.A neural network is constructed based on the principle of autoencoder.A large amount of unlabeled data is added to a small amount of labeled data to train the neural network model which can be used for microseismic location.Experiments on natural seismic data set and actual microseismic data set show that semi-supervised learning can improve the performance of the network model under the condition of insufficient label samples,and effectively improve the accuracy of the network.The intelligent microseismic location method based on neural network only needs two steps: using convolutional neural network to identify the signal and inputting the identified signal to neural network for location.Compared with the traditional location method,it has the advantage of more concise process,which eliminating many intermediate steps.Finally,this thesis implements the intelligent operation of microseismic location.