Prediction Of NMR T₂ Spectrum And T₂ Cut-off Values With Machine Learning Algorithm

In recent years,deep learning algorithm is widely used to complete and generate logging data.In this study,we analyzed the mapping relationship between conventional logging data and nuclear magnetic resonance（NMR）T₂spectrum to achieve the reconstruction of NMR T₂spectrum with Long Short-Term Memory（LSTM）network model.On the meantime,we combined the T₂spectrum obtained from the same NMR core analysis experiment and the T₂cut-off data to establish a data set,realized the automatic pick-up of the T₂cut-off value of the underground NMR through training the parameters of the LSTM network model.The T₂spectrum obtained by NMR logging at each depth shows morphological changes through different distribution points with different time series,and LSTM can well control the influence of neurons at different depths on the distribution points of T₂spectrum.This thesis select logging data and petrophysical data from A oilfield offshore China as examples to test this method.Firstly,we used grey relational degree algorithm to analyze the correlation between T₂geometric mean and conventional logging curve.Then select logging curves whose correlation is higher than the set value and normalized scale it between 0 and 1 as the input of LSTM model to predict the morphological distribution of T₂spectrum in the same formation.After the T₂spectrum reconstruction,we used the T₂cut-off value of core data as a label to predict the T₂cut-off value of the whole well,compared the T₂spectrum obtained by software with the predicted results,and analyzed the reasons for the difference between them.The results show that the data predicted by LSTM neural network model is highly consistent with the actual formation data.