Method for predicting permeability of porous medium based on LSTM (Long Short Term Memory)

Abstract

The invention discloses a method for predicting the permeability of a porous medium based on LSTM (Long Short Term Memory). The method comprises the following steps: firstly, calculating the permeability of a porous medium by utilizing a computational fluid mechanics method and using the permeability as a label value of a training sample; segmenting the porous medium, extracting geometric image features of each slice, and forming a one-dimensional feature sequence; and splicing the feature sequences into a two-dimensional matrix as the input of a long-short term memory (LSTM) neural network, and predicting the permeability of a test sample by the network after training and parameter adjustment. Neural network training and testing are carried out through a three-dimensional porous medium sample, and the permeability of the three-dimensional porous medium can be well predicted through the method. The method has the biggest advantages that the porous medium three-dimensional pore structure data is greatly compressed, the compressed data preserves the spatial seriality, and then the accurate prediction of the three-dimensional porous medium permeability is realized by utilizing the sequence processing capability of deep learning LSTM (Long Short Term Memory).

Description

technical field [0001] The invention relates to the fields of hydraulics, hydrogeology, and deep learning, in particular to a method for predicting the permeability of porous media based on a long-short-term memory neural network (LSTM). Background technique [0002] Estimation of permeability in porous media is a key issue in areas such as groundwater, oil extraction, underground carbon dioxide storage, and nuclear waste leakage. The traditional method for permeability estimation is the Darcy pressure gradient method. However, the Darcy experiment has a long test cycle and high cost, and small changes in test conditions often lead to large errors in measurement data. According to Darcy's law, the permeability of porous media is determined by the geometry of the pores. Therefore, mapping pore structure to permeability can be transformed into a supervised learning problem. Many researchers have used machine learning regression methods to build surrogate models for the purp...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the deficiencies in the prior art, the present invention provides a method for predicting the permeability of porous media with a relatively simple method while maintaining high accuracy, which effectively overcomes the difficulty in training the current application of deep learning convolutional neural networks and the structural characteristics of porous media. Extract complex problems, and the sample data volume of this method can be greatly reduced

Images