GAN River-I: A process-based low NTG meandering reservoir model dataset for machine learning studies.

The GAN River-I data set is designed to provide a stern test for machine learning and geostatistical tools that wish to recreate the complex geometries of realistic facies distributions in subsurface reservoirs. It provides more complex, non-stationary facies distributions than previous open data sets, some of which have modelled channels but do not include the number and complex association of facies types of this data set. GAN River-I is a dataset of 2D layers of 3D facies models produced from a process-based simulator of a meandering fluvial system. The data set contains 25 simulated 3D cubes, converted into three datasets consisting of 16,000 2D models/images, each representing the increasing complexity of the modelled facies. The number of facies decreases between the three data sets, with nine facies, seven facies and three facies, respectively. The facies reduction is carried out by amalgamating similar facies in terms of their likely permeability and sedimentary relationships to represent flowing units in a subsurface reservoir. The data is therefore provided to allow users to increase the model complexity in a manageable and comparable way between groups using the data. GAN River-I covers a range of low NTG meandering patterns with varied avulsion rates. Each dataset comprises an ensemble of meandering models representing various plausible patterns and, therefore, can be used as a geologically plausible benchmark for testing generative models' performance. We provide three data file formats, including image, Ndarray and GSLIB, to adapt to different researchers' preferences.

Estimating the asphaltene critical nanoaggregation concentration region using ultrasonic measurements and Bayesian inference.

Bayesian inference and ultrasonic velocity have been used to estimate the self-association concentration of the asphaltenes in toluene using a changepoint regression model. The estimated values agree with the literature information and indicate that a lower abundance of the longer side-chains can cause an earlier onset of asphaltene self-association. Asphaltenes constitute the heaviest and most complicated fraction of crude petroleum and include a surface-active sub-fraction. When present above a critical concentration in pure solvent, asphaltene "monomers" self-associate and form nanoaggregates. Asphaltene nanoaggregates are thought to play a significant role during the remediation of petroleum spills and seeps. When mixed with water, petroleum becomes expensive to remove from the water column by conventional methods. The main reason of this difficulty is the presence of highly surface-active asphaltenes in petroleum. The nanoaggregates are thought to surround the water droplets, making the water-in-oil emulsions extremely stable. Due to their molecular complexity, modelling the self-association of the asphaltenes can be a very computationally-intensive task and has mostly been approached by molecular dynamic simulations. Our approach allows the use of literature and experimental data to estimate the nanoaggregation and its credible intervals. It has a low computational cost and can also be used for other analytical/experimental methods probing a changepoint in the molecular association behaviour.