Model Predictive Control for Zero Emission Industrial Processes

Fossil fuels such as oil and gas play a crucial role in global energy demand (accounting for over 80% of the world’s energy supply), and more than a quarter of today’s oil and gas supply is produced offshore. As hydrocarbon reservoirs in the Danish North Sea mature, the maximization of recovery from these mature reservoirs and the management of discharged liquids have become the main focus of offshore oil and gas production. The produced water contains oil and other pollutants and must be cleaned before it is discharged to the surrounding sea. In addition, Oil and gas consumption is also a significant source of CO2 emissions. The cement production, however, also contributes significantly to global carbon emissions, accounting for around 30%
The subject of this thesis is model predictive control (MPC) for optimal operation and zero-emission/discharge of industrial processes. Specifically, we develop real-time advanced process control (RT-APC) solutions based on the MPC algorithms for 1) the optimal operation and Oil-in-Water (OiW) control of the offshore produced water treatment system and 2) the optimal control of the cement raw-mix blending process. This PhD thesis aims to test and validate the proposed MPC and RT-APC solutions so they can be used in practice.
In this work, we 1) develop a linear-quadratic (LQ) discretization toolbox for the numerical discretization of linear-quadratic optimal control problems (LQ-OCPs) with time delays, 2) develop a CT-LMPC toolbox for the design, development and discrete-time implementation of continuous-time-designed linear model predictive control (LMPC), 3) describe the RT-APC system framework for industrial process control, 4) develop a CT-LMPC based RT-APC software application for the optimal operation and OiW control of the offshore PWT system. In the LQ discretization toolbox, we introduce three numerical methods for solving differential equation systems associated with the discrete-time system matrices of LQ-OCPs. In addition, we describe the distribution of stochastic costs of stochastic LQ-OCPs. We apply an explicit-explicit Euler-Maruyama (EM) discretization scheme to discretize the stochastic cost function and provide the numerical approximation of its mean and variance. The CT-LMPC formulation consists of an estimator and a regulator. The estimator applies a discretetime linear Kalman filter with memory to handle missing and delayed observations. The regulator involves a continuous-time designed objective function, containing 1) the reference tracking objective, 2) the input regularization objective, 3) the input rate-of-movement (ROM) objective, 4) the economics objective, and 5) the soft constraint penalty function. We apply the LQ discretization toolbox to discretize the continuous-time-designed objective function for the discrete-time implementation of CT-LMPC. We describe the RT-APC system framework for industrial process control. The RT-APC system consists of 1) the real-time simulator module for real-time simulation of the process plant, 2) the open-platform-communication (OPC) module for the data communication between the RT-APC system and the process plant, 3) the APC database module for data storage and sharing data between various modules, 4) the APC process controller module for computing the optimal control solutions and prediction of the system’s future trajectory, and 5) the user interface module for data visualization and controller tuning. Additionally, we present a case study of a simulated cement raw-mix blending process to test and validate the proposed RT-APC system. We apply the mass balance modeling methodology to describe the models of the gravity separator (GS) and de-oiling hydrocyclones (HC). We present step response experiments for the input-output model identification around an optimal steady state operation point. The optimal steady state operation point is obtained by solving an optimal steady state operation optimal control problem (OCP). We introduce an input-regularization based control strategy and compare it with the conventional fixed set-point control strategy. We apply the CT-LMPC toolbox to develop two MPCs based on the two control strategies. We present numerical experiments to test and compare the two MPCs. The results indicate that the input-regularization based MPC has a much lower variance in oil and gas production and achieves a higher profit than the conventional control strategy based MPC while keeping all operational variables within a safety range. We develop an RT-APC software application with the input-regularization based MPC solution. We test the RT-APC software application on a simulated offshore PWT system in real time. The real-time simulation results demonstrate that the RT-APC software application can control the simulated offshore PWT system without violating the OiW concentration constraint.
This thesis consists of an extended summary report and a collection of seven research papers. Five papers are published or accepted for publication in conference proceedings.