Prediction of Pressure Drop in Horizontal and Near-Horizontal Multiphase Flow using Group Method of Data Handling (GMDH) approach with the aim of reducing the curse of dimensionality; A Comparative Study

An accurate prediction on the value of pressure drop during a multiphase flow in pipelines is greatly in need in petroleum industry. Back to 1967, the first empirical correlation was developed to predict the pressure drop in pipelines. Since then, it attracts the interest of many researchers to conduct rigorous studies on this matter. However, the correlations and models that are being used in the petroleum industry nowadays seem to be out dated. At most of the time, it tends to under predict and over predict the pressure as all the correlation have superior relation only with the data used in their experiments.
The objective of this study is to construct a model with high accuracy and low complexity, by utilizing Group Method of Data Handling (GMDH) approach. Parameters that govern the pressure drop are studied to understand their significance towards the prediction of pressure drop. Once all the parameters are outlined, a model is developed and is expected to be generalized, where it can be applied in any behavior of multiphase given. GMDH approach is well known for its ability to model the relation between multiple input parameters and an output with the mean of self-organizing. Stopping criterion will be set optimally to ensure that the model will result in accurate prediction. To achieve this, MATLAB Software will be used for coding and simulation and all the results will be further evaluated in Microsoft Excel software.
The result possess by GMDH model generated in this study will be compared with Beggs and Brill correlation, Gomez et al. correlation and Xiao et al. mechanistic model as these models are the mostly applied methods to predict pressure drop for horizontal and near-horizontal conditions.
From this study, the model generated is very successful in predicting the pressure drop in pipeline where it possess the lowest Average Absolute Percentage Error (AAPE) of 12% compared to other correlation or model. Trend analysis and statistical analysis were conducted to confirm the validity of this model.
The author believes that the model generated in this study will be able to predict the pressure drop in much convenient way in petroleum industry.