Research And Application Of Coal Seam Small Fault Identification Method Based On Deep Learning

Small fault in coal seam is one of the important factors to induce coal mine safety accidents such as water inrush and rock burst.Accurately predicting small fault in coal seam is an urgent need for efficient and safe production in coal mine.The conventional seismic data interpretation methods sometimes miss the interpretation of the fault with a drop greater than 5m,and even the interpretation of the fault with a drop greater than10 m May be biased.Therefore,it is necessary to explore new technical methods to interpret the small fault with 3-5 m drop in coal seam.This thesis carried out research on intelligent fine interpretation method of small faults of coal seam based on deep learning,and realized intelligent fine interpretation of small faults of coal seam with main research contents such as construction of sample set of typical small faults of coal seam,training and testing of deep learning network,identification of small faults of coal seam with actual seismic data.In the aspect of the construction of small fault data set of coal seam,the fault samples are few and the small faults of different coal seams are different.In this thesis,a total of 5000 pairs of fault data sets were generated by using three methods: forward seismic simulation based on finite difference method,manual annotation of interpreted coal seam seismic profiles,and automatic generation based on Python platform,among which 4000 pairs were divided into training sets,800 pairs into verification sets,and200 pairs into test sets.In the aspect of deep learning network training and testing,the generated data set is used to train CNN network and U-net network respectively,and the CNN network and U-net network trained in different rounds are compared.According to the network evaluation index,U-net fault identification effect is obviously better than CNN.The Unet network is improved and the Attention mechanism is added to U-net to get attentionunet.By comparing the fault identification effect of CNN,U-net and Attention-Unet on the test set data,it is found that Attention-unet has a better effect in identifying small faults in coal seams.Its characterization of faults is more continuous and clear,and the shape and distribution of faults are basically consistent with the development of faults.It can be used as a network model for identifying small faults in coal seams based on actual seismic data.Based on the deep learning network model,the identification of small faults in 8coal seam of a coal mine in Huaibei,Anhui Province was carried out.The interpretation results of deep learning small faults basically covered the achievements of attribute interpretation and manual interpretation,and 41 new small faults were identified.By comparing the identification results of small faults in a certain working face with the actual small faults exposed by roadway and stoping face,it is found that the recognition accuracy of deep learning reaches 92.3%,which proves the good effect of deep learning in the identification of small faults in coal seams.In conclusion,deep learning can effectively improve the ability and accuracy of coal seam small fault interpretation,and has positive guiding significance and practical value for ensuring coal mine safety production.