A novel method for accurate pressure drop prediction in horizontal and near horizontal pipes using adaptive neuro fuzzy inference system based model

Abstract

Effective flow line and piping network design depends on the accurate prediction of pressure drop in multiphase flow for horizontal and near horizontal pipes. Since early 1950, several empirical correlations and mechanistic models have been developed to predict pressure drop. All correlations used by the industry, in addition to their applicability limitations, fall short of providing the necessary precision of pressure drop predictions. However, compared to empirical correlations, the recently developed mechanistic models improved pressure drop prediction. To design and construct more dependable and economical surface piping networks and wells, it is still necessary to improve prediction accuracy. This study uses the Adaptive Neuro-Fuzzy Inference System (ANFIS) to create a model that predicts pressure drop in horizontal and near-horizontal pipelines with greater accuracy and simplicity. Using the ANFIS method, the fuzzy modelling procedure can gather knowledge about a set of data to determine the membership function parameters that will enable the associated fuzzy inference system to track input/output data most effectively. The model was created and tested using field data encompassing various variables. The model was developed using 450 different data sets that were collected from the Asian continent. 113 data sets were used for testing, and a total of 337 data sets were used for training. Trend analysis was carried out during the model development phase prior to the model’s completion. This is performed to make sure the model is stable and to make sure the created model is physically sound and accurately simulates the real physical process. To determine the percentage of error between the predicted value and the actual measured data, statistical analysis was carried out. To compare the performance of the new ANFIS model to earlier empirical correlations and mechanistic models, graphical and statistical techniques were also used. The new model outperformed known correlations and the most recent mechanistic models by a significant margin in producing incredibly accurate pressure drop predictions. The Dukler et al. empirical correlation, Beggs and Brill empirical correlation, Xiao mechanistic model, and Gomez mechanistic model had values of 25.284, 20.940, 30.122, and 20.817, respectively, while the ANFIS model had a value of 13.256 for the lowest average absolute percentage error. Additionally, the Duckler and Beggs & Brill models came in second and third, with values of 0.908 and 0.906, respectively, and the ANFIS model had the highest coefficient of determination at 0.955.

Keywords

• Artificial intelligence
• Pressure drop
• Multiphase
• Fuzzy Logic
• Neural Networks