| Nuclear magnetic resonance logging is an important logging method.Nuclear magnetic resonance logging can measure some petrophysical parameters that cannot be obtained by conventional logging methods,and plays an important role in the exploration and development of oil and gas resources.In recent years,with the development of deep learning,more and more research results combining deep learning and well logging have emerged in the field of well logging.Deep learning has been widely used in the field of logging,and its application effect is also significant.In this thesis,the inversion of nuclear magnetic resonance logging is carried out based on deep learning.In this thesis,three different network models are used to invert the NMR T2distribution.The inversion results of the model are verified on the generated forward model data set and the measured data,and good results are obtained.At the same time,this thesis proposes two methods to improve the network model according to the characteristics of the input data.In the echo string obtained by NMR measurement,the effective information is retained at the front end of the echo string,and the first wave amplitude of the echo string is approximately equal to the NMR porosity.Based on the above principles,this thesis adds the Attention layer to the network,and adds the first wave amplitude as a new input layer to the model.The improved model achieves better inversion accuracy.Through a series of experiments,this thesis verifies that a variety of network models based on deep learning have better inversion results than traditional inversion methods,and also have the conditions to be deployed in actual well logging projects. |
