Plantwide control of an oil production network

Esmaeil Jahanshahi, Dinesh Krishnamoorthy, Andrés Codas, Bjarne Foss, Sigurd Skogestad
2020 Computers and Chemical Engineering  
In this paper, we consider Real-Time Optimization (RTO) and control of an oil production system. We follow a systematic plantwide control procedure. The process consists of two gas-lift oil wells connected to a pipeline-riser system, and a separator at the topside platform. When the gas injection rates are low, the desired steady flow regime may become unstable and change to slug flow due to the casing-heading phenomenon. Therefore, a regulatory control layer is required to stabilize the
more » ... tabilize the desired two-phase flow regime. To this end, we propose a new control structure using two pressure measurements, one at the well-head and one at the annulus. For the optimization layer, we compare the performance of nonlinear Economic Model Predictive Control (EMPC), dynamic Feedback-RTO (FRTO) and Self-Optimizing Control (SOC). Based on dynamic simulations using the realistic OLGA simulator, we find that SOC is the most practical approach. (S. Skogestad). be used for stabilization while slower centralized optimizers may be used for long-term optimization ( Skogestad, 2004 ) . In our study, we initially attempted to solve an optimal oil production control problem using a centralized NMPC approach, that is, with a single optimizing controller. However, we were not successful. This was partly because the plant is unstable, also because of the plant-model mismatch. We used the Olga simulator as the "real" process and a simplified dynamic model for the control design. Willersrud et al. (2013) successfully applied a centralized control structure (NMPC) to another oil and gas production system. However, they assumed no model-plant mismatch; that is, they considered the same model for both optimization and simulation. Thus, robustness against modeling errors was neglected. Moreover, this single-layer centralized strategy was not tested in closed-loop with unmeasured disturbances. Complex industrial processes require a structured control architecture for their operation. Skogestad (2004) ; Saputelli et al. (2006) ; Foss (2012) propose to decompose the control and optimization problem on different time scales. Luyben et al. (1997) and Skogestad (2004) propose systematic procedures for design of such plantwide control systems. The resulting multi-layer plantwide control structure in Fig. 1 ( Skogestad, 2004 ) is well established in the process industry, see e.g. , Campos et al. (2015) . The lower control layers are fast and do not affect the optimization of the process. In practice, the slow Real-Time Optimization (RTO) layer is designed based
doi:10.1016/j.compchemeng.2020.106765 fatcat:n2ziprfmgvdpvn2hb2h2zswfuy