Precise estimation of the effective petrophysical characteristics for the oil or gas-bearing reservoir plays a vital role in production control problems. It is preferable to use nondestructive measurement methods for porosity, conductivity, geomechanical modulus, and some other parameter fields. For the estimation of anisotropic permeability tensor flows in different directions need to be simulated in the core plug digital image void space, which is very difficult (if possible) to conduct with the real rock sample. The creation of the digital core image includes three stages: construction of its internal structure based on the computed tomography (CT) sinogram, filtration, and segmentation. Routine practice is the filtration of 2D slices pack of the core plug CT image because of a lack of computational power and memory limitations. Unfortunately, this can generate a directional orientation error orthogonal to the pack of slices. The total 3D edge-preserving filtration with the usage of two approaches: modified implicit anisotropic diffusion and discrete orthogonal transforms can reduce this error. Universal code based on the MPI+OpenAcc programming paradigm was tested on different high-performance computing systems incorporating various accelerators like GPGPU and heterogeneous processors such as Sunway 26010.

No abstract

We present a new regularization method called Weights Reset, which includes periodically resetting a random portion of layer weights during the training process using predefined probability distributions. This technique was applied and tested on several popular classification datasets, Caltech-101, CIFAR-100 and Imagenette. We compare these results with other traditional regularization methods. The subsequent test results demonstrate that the Weights Reset method is competitive, achieving the best performance on Imagenette dataset and the challenging and unbalanced Caltech-101 dataset. This method also has sufficient potential to prevent vanishing and exploding gradients. However, this analysis is of a brief nature. Further comprehensive studies are needed in order to gain a deep understanding of the computing potential and limitations of the Weights Reset method. The observed results show that the Weights Reset method can be estimated as an effective extension of the traditional regularization methods and can help to improve model performance and generalization.

Lack of petrophysical information is critical for reservoirs development composed of poorly consolidated rocks or for zones bearing wells with core damaged by improper coring operations. The restoration complexity of the digital-core lost sections is associated with the need to consider an enormous amount of data from the existing core image and the necessity to include lithological expert knowledge. That makes deep learning methods a natural choice for solving such problems. We proposed, examined, and compared several deep learning methods convenient for analyzing micro-computed tomography digital core data. It was done under the most simplistic problem statement when the destroyed part (a set of slices) is completely lost. Here, we present the results of comparison interpolation/extrapolation procedures under proposed quality metrics. We discover that the variational autoencoder method can be trained to extract some petrophysical parameters from the digital core plug in an unsupervised manner.