The Reservoir Permeability Prediction Based On Machine Learning And NMR Logging Data

Reservoir permeability information is one of the important parameters related to reservoir exploration and development,and has important guiding significance for drilling,completion,perforation and other follow-up work.Traditional permeability calculation methods in NMR logging field need to calculate by existing models,so it is difficult to accurately calculate the permeability of rocks with complex structures.Machine learning can use data to build statistical models that learn from the data and use the models learned from the data to analyze and predict new data.In this thesis,the core automatic positioning is firstly processed by correlation analysis,and the input echo data is sparse extracted.This thesis designed the depth neural network,the convolutional neural network and long-short term memory neural network of three network model.And established the nuclear magnetic resonance（NMR）in three models respectively T₂distribution,nuclear magnetic resonance（NMR）echo data string of data and the relationship between the permeability and permeability tests prove that when the long-short term memory neural network to forecast effect is best.Due to different lithology reservoir to the different characteristics,different lithologic reservoir,this thesis has carried on the lithology prediction processing,respectively using depth neural network and the convolutional neural network on the two data points lithology prediction,because this article data quantity is less,the result is relatively poor,and easy to fall into local optimum in the training process.In order to further improve the accuracy of prediction,NMR logging data were processed by feature engineering.In view of the large dimension of NMR data,principal component analysis and linear discriminant analysis were used to reduce the dimension of the data.LDA-LSTM model and CNN-LSTM model were designed to achieve dimensionality reduction prediction of NMR data.Both methods improved the prediction permeability of pure LSTM model,especially for the input NMR echo string data,and MAE was increased by more than 32%.This provides a basis for making full use of logging data,helps accurately predict reservoir parameters,and has guiding significance for oil and gas field development.