Imaging Coal Seam Anomalies Based On Elastic Full Waveform Inversion

The coal industry is an important pillar of our national economy,and the coal consumption accounts for over 60% of the disposable energy structure.The geological conditions are related to coal mine safety production,and geological anomalies in coal seams are direct factors that induce coal mine safety accidents.With years of coal production,the coal mining conditions have become increasingly complex and difficult.Conventional seismic exploration methods are hard to meet the requirement for efficient and safe mining.Therefor,it is of great significance to carry out the research of high precision and high resolution exploration methods for fine reconnaissance in underground coal mine.In-seam seismic exploration is a relatively mature method,which has been widely used in coal seam reconnaissance.Although the seam wave exploration has great advantages and precision over the other methods,but it behaves poor in detecting small scale anomalies and other abnormal bodies in coal seam.The seismic full waveform inversion,as a high precision and high resolution inversion method for physical parameters of underground media,is one of the hotpots in the field of seismic imaging in recent years.Full waveform inversion takes full advantage of all kinematic and dynamic information in seismic records,and thus has the ability to depict complex geological structures.In order to solve the problem of fine detection in underground coal mines,the full waveform inversion method is introduced to explore the feasibility and adptability of inversion under the condition of coal seam wave.Through detail researches in the main aspects including finite difference method in forward modelling,free boundary condition,the basic theory of elastic full waveform inversion,gradient calculation,step length estimation,GPU parallel computing,inversion of geological anormalies in coal seam,the following results are obtained:(1)The traditional 2D undulating surface mesh classification scheme are improved to realize the 3D omni-directional grid point classification scheme in tunnel or roadway,which solves the problem of the surface wave of roadway or tunnel.(2)A scheme of 2D elastic full waveform inversion based on a single GPU card is designed.The wavefield reconstruction method is adopted to solve the problem of insufficient global memory on GPU;By comparing the computational efficiency of different thread dimension configurations,the key factors that affect the computation efficiency of parallel algorithms are found;By comparing the algorithms based on CPU parallel and GPU parallel,the parallel acceleration ratio of single GPU card to single CPU and eight thread CPU forward calculation and gradient calculation is obtained respectively.(3)A 3D elastic wave full waveform inversion scheme based on multi node and multi GPU is designed.A parallel strategy based on model domain decomposition is implemented.Each sub-model is calculated by a single GPU card,and the data communication between the sub-models is carried out by MPI,thus the dual parallelism of the whole inversion process is realized;A parallel computing strategy for computing 3D model based on 2D thread grids is implemented to solve the problem of insuffienct shared memory in thread blocks;By first calculating the boundary area and then the inner region of the submodel,multiple CUDA streams are used together to hide the exchange time of boundary area data of the submodel between the nodes in the calculation,thus the coarse granularity parallel on GPU is realized;Through the tests of multi GPU parallel computing and single GPU parallel,the parallel acceleration efficiency of multi GPU compared to single GPU is obtained.(4)The numerical inversion results of the common coal seam anormalies show that the full waveform inversion has good adaptability and stability to the seam wave,and the inversion can restore the collapse column,fault block,dirt bands and the undulated coal seam,which achieve a high resolution.The contrast of the inversion results shows that because of the existence of coal seam,the propagation of seismic waves will be constrained in the seam,the anormalies in the coal seam can be retrieved more accurately,while the anormalies outside the coal seam show relatively poor resolution.(5)The acquisition geometry has an effect on the results of full waveform inversion.Through the study of the influence of the layout of different observation systems in the coal seam on the inversion results,it reveals that if we want to obtain better inversion results and ensure better vertical(coal thickness)resolution,it is necessary to set up several shot line and receiver line in the vertical direction.For short,the more shots and receivers,the better effect the inversion has.The highlights of this dissertation:(1)The full waveform inversion method is introduced to the reconnaissance of geological anomalies under the condition of coal seam wave.The inversion results of the numerical model of common abnormal body including fault,dirt bands,collapse column,undulated country rocks in coal seam show that the full waveform inversion can adapt to the condition of the seam wave in the coal seam,and the method is capable of depicting the small scale structure more accurately.Full waveform inversion method can provide a new solution for the fine seismic exploration in coal seam.(2)The traditional surface free surface grid point classification scheme has been developed so that it can be applied to 3D tunnel or roadway model.(3)A parallel implementation scheme of 3D elastic wave full waveform inversion based on multi node and multi GPU is designed.Some strategies are realized including thread level parallel and task level parallel,thread block optimization,utilization of shared memory,register and other resources,which take full advantage of the GPU performance and improve the computational efficiency of the 3D inversion problems.