Predictive-adaptative production control in hydrocarbonreservoirs

Abstract

This paper presents a predictive-adaptive strategy for the control of the total production at the end of a production horizon, in a reservoir with multiple injecting and producing wells. This strategy includes, at each sampling instant, aspects such as: i) identification and continuous update of a neural network-based ARMAX model (NN-ARMAX) for each producing well, (ii)  linearization of each NN-ARMAX model at the current operating point of the reservoir, (iii) construction of a linear MIMO Kalman  innovation model from the linear ARMAX models, iv) calculation of the injection rates using a predictive control scheme and v) implementation of the injection rates in the reservoir. In a waterflooding recovery process, the proposed strategy compares favorably against an empiric strategy of constant injection rate and a decentralized PID control strategy, for two reservoirs models with different levels of heterogeneity and different arrays of injecting and producing wells, under different performance measures such as time to reach the production set-point for each well, mean square error on the production for each well, relative error in the total production and volumes of injected and produced water. Â

Keywords

• Predictive-adaptive control of oil production
• EOR
• Smart fields
• Data mining