Research On Prediction Method Of Mine Earthquake In Time Sequence Based On Clustering Analysis

Coal still plays a dominant role in the energy system of our country,which will not change in the short term.The high-energy mine earthquake caused by its mining will not only induce underground rock burst disaster,causing casualties and property losses,but even lead to the destruction of ground buildings,causing panic of public opinion.In order to reduce the occurrence of coal mine dynamic disasters and reduce casualties and property losses,it is necessary to carry out monitoring and early warning research for high-energy mine earthquake.Microseism monitoring system is very effective in mine microseism monitoring and has been widely used.It can realize long-distance,real-time,dynamic and automatic monitoring of mine earthquake,and give the complete waveform of mine earthquake signal.Each data record in the system contains the time,space and intensity information of an event.Compared with the traditional experience-driven and mechanism-driven mining earthquake prediction methods,the data-driven prediction method can more fully mine the effective information of massive microseism data.At the same time,mine microseisms have significant spatiotemporal aggregation,and the clustering of microseism events can improve the prediction effect of the subsequent prediction model on high-energy mine earthquakes.Therefore,a time series prediction method of mine earthquakes based on cluster analysis is proposed in this thesis to predict large energy mine earthquakes with disaster potential.Specific research contents are as follows:(1)Microseism data is spatiotemporal data,so the spatiotemporal density clustering algorithm(ST-DBSCAN)is used to identify the active areas of microseisms on the mining face,and the microseism events with close spatiotemporal correlation are divided into the same clusters,and the contour coefficient and noise rate are selected as the evaluation indexes of clustering effect.After clustering,the regularity of time series of microseism events in the same cluster is strengthened compared with that before clustering,which is helpful to improve the effect of subsequent prediction model.(2)Microseism events on the entire mining working surface are a large sample data set,while microseism events contained in a single cluster after clustering are a small sample data set.The support vector regression method(SVR)has significant advantages in the time series prediction of small sample data.Therefore,it is used to establish a time series prediction model to predict the occurrence location and energy level frequency of high-energy mine earthquakes,and the mean square error(MSE)and goodness of fit(R~2)are selected as the evaluation indexes of prediction effect.(3)As comparative experiments,on the whole working face microseism data set without clustering processing,SVR and LSTM,which have advantages in time series prediction of large sample data,are used to establish prediction models,respectively,to predict the occurrence location and energy level frequency of high-energy mine earthquakes.The purpose of the comparison experiments is divided into two levels.On the one hand,it shows that the prediction effect of SVR model is improved by using the microseism data after clustering processing;On the other hand,it shows that the time series prediction model combining ST-DBSCAN and SVR has a better effect on the prediction of large energy mine earthquakes than the prediction model established directly using LSTM.Comparative experiments show that the data driven method proposed in this thesis,which combines unsupervised machine learning with supervised machine learning,is effective in mining earthquake prediction.The thesis has 30 pictures,11 tables and 97 references.