Improved predictions of wellhead choke liquid critical-flow rates: Modelling based on hybrid neural network training learning based optimization

Abstract Published relationships typically consider liquid critical-flow rate through wellhead chokes of producing oil wells as functions of wellhead pressure, choke size and gas-liquid ratio. Such correlations can be improved by taking into account three additional input variables: gas specific gravity, oil specific gravity and temperature. Novel liquid critical-flow rate models, hybridizing an artificial neural network (ANN) with a teaching-learning-based optimization (TLBO) algorithms, involving 3 and 6 input variables, demonstrate improved accuracy compared to nonlinear regression models, traditional ANN models and published correlations. The improved accuracy of the developed models is assessed statistically using a data set of 113 wellhead flow tests from oil wells in South Iran (with a full data listing included). The ANN-TLBO (6 parameters) developed model is the most accurate, yielding the best liquid critical-flow rate predictions for that data set: coefficient of determination of 0.981; root mean square error of 714; average relative error of 2.09%; and, average absolute relative error of 6.5%. The 6-parameters models outperform the 3-parameters models without over complicating model functionality. This justifies the consideration of all six input variables to deliver improved predictions of wellhead choke liquid critical-flow rates. Calculation of relevancy factors for the 6-parameters ANN-TLBO model to the data set for all six input variables reveals choke size and gas-liquid ratio have maximum and minimum influence in determining the liquid critical-flow rate, respectively.