ABSTRACT:

Hydraulic fracturing has received increasing attention in recent years for its ability to improve unconventional oil and gas production rates and recovery. The migration of solid particles in fractures has become an important process such as the placement of proppants and the control of irregular formation sand, while the existing studies have been limited to spherical particles. Perfect prediction of settling velocity of arbitrarily shaped particles in fractures is conducive to accurate calculation of proppant transport distance and productivity threshold for the control of sand. With the influence of irregular shape and the wall effect, the settling behavior of particles in fracture is complex to be described accurately by the traditional data fitting methods. Artificial neural network (ANN) is a biological inspired machine learning method, which has significant advantages in solving complex multi-parameter and nonlinear problems. In this study, ANN is used to determine the relationship between variables such as particle properties, fluid properties and wall effect and settling velocity. The dataset matrix involving 25 kinds of particle shapes included the particle Sphericity of 0.47-1.00, the dimensionless diameter of 0.01-0.95, particle Reynolds number range of 0.01-482.67. The performance of Sphericity, Circularity and Corey factor on describing the settling behavior of arbitrary shape particles were compared and analyzed, an ANN models of settling velocity with mean squared error (MSE) of 0.0067 and mean absolute error (MAE) of 0.0345 for arbitrary particle in vertical fracture was developed. In addition, the weight coefficient method is introduced to analyze the relative importance of different variables to the settling velocity. For the repeatability of the model, the weights and thresholds of the neural network model with are provided with mathematical expression. This study is of great significance to improve the parameters of hydraulic fracturing and dynamic control of formation sand.

This content is only available via PDF.
You can access this article if you purchase or spend a download.