Artificial neural network (ANN) based prediction of proppant settling in horizontal wellbores during hydraulic...

Proppant transport and settling in horizontal wellbores is a major challenge in hydraulic fracturing, leading to problems such as sand production, equipment wear, wellbore blockages, and reduced production rates. Traditional empirical models are often limited in accuracy because the physical relationships involved are highly nonlinear and complex. In this study, Artificial Neural Networks (ANNs) were used to develop predictive models for sand settling in horizontal wellbores during hydraulic fracturing. In this work, the data were obtained from controlled laboratory experiments that simulated horizontal wellbore sections with different perforation clusters. Key parameters such as proppant diameter, injection rate, perforation orientation, and number of perforations were analyzed. Two ANN models were developed: Model A used all nine measured parameters, while Model B used five selected parameters identified through correlation analysis. Model performance was evaluated using statistical metrics including the coefficient of determination (R²), Average Absolute Difference (AAD), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). Additionally, comparative analyses with Random Forest and Gradient Boosting algorithms confirmed the superior performance of the ANN models, and an explicit neural-network-based correlation was formulated for direct engineering use. Results show that Model A achieved an R² of 0.96 and Model B achieved 0.89, demonstrating that input reduction only slightly reduced predictive accuracy. Both models successfully captured nonlinear relationships, confirming that injection rate, perforation orientation, and perforation number are the most critical factors influencing sand settling. To further test their robustness, cross-validation was carried out using independent experimental data from the literature. Model A achieved an R² of 0.82 and Model B achieved 0.75, showing that both models generalized well to independent datasets, with Model A slightly outperforming Model B. This study provides a novel machine-learning approach for predicting proppant settling in horizontal wellbores under fracturing conditions from experimental data. In contrast to empirical models, the ANN-based predictor accounts for multivariable nonlinear interactions and can be deployed for real-time decision support. The findings contribute to enhanced hydraulic fracturing designs, improved wellbore stability, and reduced operational challenges related to sand production and equipment erosion. Overall, the ANN-based models provide quick and reliable predictions of sand settling, outperforming traditional empirical approaches and offering practical tools for optimizing hydraulic fracturing designs. By accounting for nonlinear interactions and validating independent data, the models demonstrate strong potential for real-time decision support, improved wellbore stability, and reduced operational challenges related to sand production and equipment erosion.