CUDNN-LSTM in Applied Geophysics: A Breakthrough in Permeability Estimation and Hydrocarbon Exploration

Prospective delineation of Permeability is an essential determinant of the productivity of an oil and gas reservoir. Precisely assessing permeability is crucial for enhancing productivity and minimizing operating expenses. This study employed the CUDNNLSTM algorithm to assess reservoir permeability. The drilling core data was partitioned into a training set and a validation set, utilizing ۸۰% of the data for training and ۲۰% for validation. In light of the significant variability in permeability across the formation, we devised the CUDNNLSTM method for permeability estimation. Initially, owing to the extensively scattered signals from the acoustic, density, and neutron logs associated with permeability, we modified the method to train for ۱۰۰۰ epochs. Upon the validation loss attaining a value of ۰.۰۱۵۸, the algorithm autonomously terminated the training process during the ۵۰۰th period. After ۵۰۰ epochs of the algorithm, we attained a remarkable accuracy of ۹۸.۴۲%. Utilizing the technique, we assessed the permeabilities of the complete array of wells, yielding extremely satisfying findings. The CUDNNLSTM method, owing to its extensive neuron count and capacity to resolve high-order equations on the GPU, serves as an effective instrument for precise permeability estimation in oil and gas reservoirs. Its capacity to manage extensively scattered signals from diverse logs renders it an invaluable resource for improving output and minimizing operating expenses, since it is far more economical than core extraction and possesses exceptional precision.

Keywords:

Permeability Estimation , Deep Learning , CUDNN-LSTM , Well Log Analysis