Controle preditivo baseado em modelo fuzzy embarcável para poços de petróleo equipados com bombeio centrífugo submerso

Abstract

Advanced control techniques have not yet been a consolidated reality in the oil and natural gas artificial lift sector, particularly in electric submersible pumping (ESP) systems. Model-based predictive controllers (MPC) require representative system models, which can be a costly task using a phenomenological approach due to the fluid-dynamic and thermodynamic complexity of an oil and gas well. Therefore, data-driven approaches are viable alternatives to achieve a balance between accuracy and time in obtaining the dynamic model of these production processes. Takagi-Sugeno-Kang (TSK) fuzzy inference systems have this characteristic, as they can approximate nonlinear dynamics through a set of linear submodels. By combining this approach with MPC, it is possible to use a simplified linear control structure that is computationally efficient, while still providing the necessary accuracy and representativeness for the process. This forms a favorable condition for use in artificial lift systems, which often have isolated installations and depend on dedicated control systems. The proposal of this work is the synthesis of a predictive controller with guaranteed feasibility, encapsulating a TSK fuzzy prediction model for ESP-type lifting systems with low computational cost. The focus is on the direct application of the technique through embedding it in microcontrollers, simulating the industrial control infrastructure. The implemented case studies on a BCS system using the hardware-in-the-loop (HIL) approach demonstrated a reduction of over 15% in control costs, an improvement of more than 10% in operational envelope constraint, and a computational time for the embedded optimization problem approximately 20 times shorter than the system’s sampling period. The representativeness of the TSK model and the control capability and feasibility of the technique were verified in the face of typical constraints and a reduced number of measured variables.