Prediction of Single Proppant Terminal Settling Velocity in High Viscosity Friction Reducers by Using Artificial Neural Networks and XGBoost

Abstract

Abstract High viscosity friction reducers (HVFRs) have been recently gaining more attention and increasing in use, not only as friction-reducing agents but also as proppant carriers. The settling velocity of the proppant is one of the key outputs to describe their proppant transport capability. However, it is influenced by many factors such as fluid properties, proppant properties, and fracture properties. Many empirical/physics-based models and correlations to predict particle settling velocity have been developed. However, they are usually based on certain assumptions and have applicable limits. In contrast, machine learning models can be considered as a black box. The objective of this study is to use machine learning models to find the relationship between the multiple factors mentioned above and particle settling velocity in order to correctly predict it. Two of the most popular and powerful machine learning algorithms, Artificial neural networks (ANN) and XGBoost, were comparatively investigated with standard data processing and training procedures. Mean Absolute Errors (MAEs) for ANNs and XGBoost were 0.010379 and 0.004253 respectively. The XGBoost learning algorithm had overall better prediction performance than the ANN model in terms of the data sets used for this study and had the potential to properly handle missing values by itself.