Otimização da produção diária orientada a dados de poços de petróleo assistidos por gas-lift

Abstract

The present study addresses a dynamic optimization problem formulation for daily oil and gas production through a data-driven approach. To this end, it is proposed to use an artificial neural network (ANN) architecture as a surrogate model for the phenomenological one, which represents a system of producing wells assisted by continuous gas injection (Gas Lift). To carry out ANN training, it is essential to have relevant data, especially concerning economic optimization. Therefore, including information about each well's oil and gas flow rates and the top of the \textit{riser} is crucial. However, flow measurements are not available in real-time, and what is available is only the total flow rate of the field, measured in the topside installations (after the three-phase separator), which does not allow understanding the behavior of each well individually. Due to the unavailability of individual well's flow rate measurement, the use of a moving horizon estimator (MHE), supported by a phenomenological model, proved to be an appropriate solution to estimate these variables, allowing data to be made available for training and obtained from a substitute model to be used in the dynamic optimization stage. In this way, it was possible to train an ANN with the \textit{ Nonlinear Autoregressive with Exogenous Input} architecture used in this work. This choice was made based on the observation that the network could predict one ahead step (one day) conveniently. The application results in a well, and a field composed of three wells and a \textit{riser} showed a good performance of the artificial neural network in terms of temporal prediction, in addition to presenting a more efficient computational time in solving the optimization problem when compared to the standard phenomenological model. The solution proposed by this approach opens up possibilities for implementation in large-scale problems, such as optimizing the daily production of an oil field composed of several wells integrated by different reservoirs and manifolds.