The focus of this paper is on the optimal control of a natural gas stabilization unit. The primary purpose of the stabilization unit is to recover the intermediate and heavy C5+ components to generate greater revenue (objective function), which can be constrained by disturbances and uncertainties due to feedstock properties, production cost and market demand variabilities. The overall objective of this work is the development of data-driven self-optimizing control of a gas stabilization unit, the performance of which was compared to open-loop optimization. Towards this objective, this paper falls mainly into four parts; namely (1) mathematical model formulation for a gas stabilization process (2) process simulation, model verification and optimization (3) data acquisition from verified model (4) developing a data driven self-optimizing control method. The first part addresses the development of a steady state mathematical model of a natural gas stabilization unit. This model presents a particularly useful way of calculating hydrocarbon Vapor-liquid equilibrium in the system. The second part involves the verification of the developed model by simulating data obtained from literature. A stabilization unit with the aid of a computer software simulation package (Aspen plus) was used to estimate material and energy balance. Based on the yields obtained from the simulation, the developed model presents reliable and promising results when compared to the simulation package. Additionally, sensitivity analysis was carried out to determine the effect of different operating-parameters on products yield and annual profit (objective function). Data acquisition from the developed stabilization model was carried out in the third part. Finally, the fourth part present a data-driven self-optimizing control method for the gas stabilization unit. A combination of measurements was used to approximate the gradient function of the process using regression technique. Three sets of polynomial were used for the regression purposes to approximate the gradient function and used as a self-optimizing control variable (CV). The measurements considered as predictors of the gradient are column temperature T, vapor flow rate V and specific gravity of oil (bottom product) from the stabilization unit. After conducting the regression, performances of different CVs were evaluated numerically using the steady-state loss function. Four scenarios were considered where each consist of 1, 2, 3 and 4 disturbances. Each scenario presents different cases with different combination of disturbance. In total 15 case model were run. The performance of each case is evaluated by comparing self-optimizing control method (SOC) with open loop optimization (OL) using average economic loss where the method with lesser percentage gives a better performance. The fourth scenario which considered all four input disturbance presents an average economic loss of 5.72% for SOC, while outperforming OL with a loss of 42.11%. The performance observed and results obtained from all cases suggest that modeling and optimizing such processes using regression models for self-optimizing control and appropriate sensitivity analysis techniques is a promising approach to natural gas stabilization problems under uncertainties. This work represents a significant progress for optimization of natural gas stabilization process operations, and shows a promising evidence that a new generation of process optimization technology based around these advances in data-driven SOC would be of immense value to the oil and gas industry.
