Prediction of Critical Multiphase Flow Through Chokes by Using A Rigorous Artificial Neural Network Method

• Published in: Flow measurement and instrumentation 2019-10, Vol.69, p.101579, Article 101579
• Main Authors: Rashid, Saeed, Ghamartale, Ali, Abbasi, Jassem, Darvish, Hoda, Tatar, Afshin
• Format: Article
• Language: English
• Subjects: Choke modeling; Flow rate; Genetic algorithm; Radial basis function
• ISSN: 0955-5986; 1873-6998
• DOI: 10.1016/j.flowmeasinst.2019.101579

Passing the flow through a choke valve is one of the most important and valuable techniques in oil production. Liquid flow rate is an important factor to assess oil wells from an operational and economic point of view. There are some validated models that predict the flow rate of single phase fluid in a wellhead condition. However, the fluid is mostly multi-phase and lies in the critical condition when passing through the choke valve. A large number of scholars have made abortive attempts to develop a universal method to predict this flow rate. Based on the available empirical equations, the liquid rate depends on upstream pressure, gas-liquid ratio, and the bin size of the choke valve. To fill the current gap, this paper seeks to develop a model that can predict the multiphase flow behavior of choke valve in critical conditions by means of Radial Basis Function (RBF) neural network coupled with Genetic Algorithm (GA), as the optimizer. The model was developed using 221 training and 55 testing data points. The obtained results were then compared with the field data and, accordingly, the eligibility of the selected method was verified. Moreover, the dependency of input parameters on the liquid rate was evaluated using the Pearson and Spearman methods to show the effectiveness of each input parameter. While upstream pressure and gas-liquid ratio showed an inverse relationship, the choke bin size showed a direct relationship with the liquid rate. •An intelligent model for prediction of multiphase critical choke flow is developed.•Statistical analysis shows acceptable predictions for a wide range of input variables.•Outlier diagnosis is performed to detect the probable erroneous measurements.•The effects of each input variable is depicted through two various sensitivity analysis techniques.