OTIMIZAÇÃO ECONÔMICA EM TEMPO REAL ORIENTADA A DADOS PARA POÇOS DE PETRÓLEO COM INJEÇÃO DE GÁS

Abstract

This work addresses the daily economic optimization problem in gas-lift oil wells. The proposal uses an artificial neural network (ANN) to find the optimal solution with a lower computational cost than one produced with a phenomenological model. However, a first-principle nonlinear model of a gas-assisted oil well production system was developed to generate the experimental data. From the practical point of view, a moving horizon estimator was designed as an adequate solution for estimating hard-to-measure variables such as produced oil and gas flow rates. Thus, it was possible to train the ANN with Nonlinear Auto-regressive with Exogenous Input-type architecture. The results showed a good ANN performance in terms of prediction capability suitable to daily optimization horizon (one-step ahead) and a better computational time in the optimization problem solution compared with the standard phenomenological model. The solution approach opens possibilities for large-scale problem implementations, such as the daily optimization of the oil field production (multiple wells integrated by reservoir and manifold).