Defesa de Dissertação de Mestrado – Jean Panaioti Jordanou – 1/8/2019

19/08/2019 14:26
Defesa de Dissertação de Mestrado
Aluno Jean Panaioti Jordanou
Orientador Prof. Eduardo Camponogara, Dr. – DAS/UFSC
Coorientador Prof. Eric Aislan Antonelo, Dr. – UEM
Data 1/8/2019 (quinta-feira) – 8h30

Sala PPGEAS I (piso superior)

Banca Prof. Eduardo Camponogara, Dr. – DAS/UFSC (presidente);

Prof. Leandro dos Santos Coelho, Dr. –PUCPR;

Prof. Rodolfo Cesar Costa Flesch, Dr. – DAS/UFSC;

Prof. Gustavo Artur de Andrade, Dr. – DAS/UFSC.

Título Echo State Networks for Online Learning Control and MPC of Unknown Dynamic Systems: application in Control of Oil Wells
Abstract: As technology advances over time, data-driven approaches become more relevant in many fields of both academia and industry, including process control. One important kind of process that benefits from data-driven modeling and control is oil and gas production, as the reservoir conditions and multiphase flow composition are not entirely known and thus hindering the synthesis of an exact physical model. With that in mind, we apply control strategies utilizing Echo State Networks (ESN) in an oil and gas production plant model. In the first application, we use an ESN to obtain online the inverse model of a system where two gas-lifted oil wells and a riser are connected by a friction-less manifold, and use the resulting model to compute a set-point tracking control action. We successfully perform setpoint tracking in three different combinations of input and output variables for the production system, some multivariate. In the second method, we train an ESN to serve as the model for a Practical Nonlinear Model Predictive Control (PNMPC) framework, whereby the ESN provides the free response by forward simulation and the forced response by linearization of the nonlinear model. The ESN is an analytical model, thus the gradients are easily provided for the linearization. The ESN-PNMPC setup succesfully performs reference tracking of a gas-lifted oil well bottom-hole pressure, while avoiding operational constraints such as saturation, rate limiting, and bounds on the well top-side pressure. This work contributes to the literature by showing that these two ESN-based control strategies are effective in complex dynamic systems, such as the oil and gas plant models, and also as a proof of concept for the ESN-PNMPC framework.