Improved Estimation of Surface Flow Line Pressure Drop Using Machine Learning Techniques

Accurate estimation of pressure drop in surface flow lines is essential for optimizing oil production and ensuring reliable surface network design. Multiphase flow behavior introduces significant complexity due to gas-liquid interactions, holdup variations, and flow-regime transitions, which can strongly influence well productivity. Conventional empirical and mechanistic correlations for pressure-drop estimation have been widely used; however, these methods are typically developed under limited laboratory or regional conditions and often fail to generalize to diverse field data. This study aims to develop and validate machine-learning-based models-specifically Artificial Neural Network (ANN) and Adaptive Neuro-Fuzzy Inference System (ANFIS)-to predict multiphase pressure drop in surface flow lines. Additionally, a new ANN-based empirical correlation is introduced for direct implementation in engineering workflows. A total of 450 field datasets were collected from surface production systems exhibiting multiphase flow. Twelve operational and fluid parameters were selected as inputs, and the measured pressure drop across the surface flow line served as the output. The dataset was divided into training, validation, and testing subsets to ensure model robustness and generalization. Both ANN and ANFIS models were developed, trained, and optimized using performance metrics such as Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). The models' predictions were compared with results obtained from standard empirical correlations available in PROSPER petroleum software. From the best-performing ANN model, a new empirical machine-learning-based correlation was derived to allow direct field application. The machine-learning models demonstrated substantially higher predictive accuracy than conventional PROSPER correlations, with both ANN and ANFIS producing lower error values and improved correlation coefficients. The new ANN-derived correlation provided reliable, simplified pressure-drop estimation without the need for iterative simulation, reducing computational time while maintaining precision. The results clearly indicate that data-driven models outperform traditional mechanistic approaches in capturing complex multiphase flow behavior under variable field conditions. This study presents a novel ANN-based empirical correlation for estimating pressure drop across surface flow lines under multiphase flow conditions. Unlike conventional correlations limited to specific operational envelopes, the new model generalizes effectively across a wide range of field conditions. The approach integrates machine learning and petroleum production modeling, aligning with the industry's digital transformation objectives. The findings offer practical guidance for engineers seeking to enhance flow assurance, improve production forecasting, and reduce operational uncertainty using intelligent data-driven tools.