An improved optimization method in gas allocation for continuous flow gas-lift system

Abstract

The primary objective in the continuous flow gas-lift operations is to inject an optimal gas volume for a group of wells to maximize oil production. Due to the gas supply constraint in an oilfield, optimization of gas injection plays an important role in achieving this goal. In this work, for modeling of gas-lift operation, the potential application of an Artificial Neural Network (ANN) using Bayesian Regularization (BR) is investigated and the results are compared with Levenberg–Marquardt (LM) back-propagation training algorithm. For the optimization, Teaching–Learning-Based Optimization (TLBO) algorithm is applied to simultaneously solve the well-rate and gas-lift allocation problems under the injection capacity constraint. The efficiency of the TLBO is investigated based on (a) convergence rate and (b) the best solution, by comparing its performance with Genetic Algorithm (GA). Extensive published data are used in model development and comparison. The proposed prediction and optimization model is tested in a gas-lift system for a given period of reservoir life. The prediction accuracy produced by the BRNN and the LMNN were 99.9% and 99.5% respectively. Results indicate that the two models have good predictive capability. Also, results show that the BR model appears more robust and efficient than the LM model and for the optimization algorithms, TBLO outperforms GA in the gas allocation mapping for continuous gas-lift system. The simulation results demonstrate the effectiveness of the proposed model on continuous flow gas-lift operations.