Proper process control enhances product yield, reduces energy consumption and improves safety, improving profitability and competitiveness. A leading technology for improving process control is multivariable model-predictive control (MPC), which identifies relationships among independent and dependent variables, and uses matrix-based control and optimization algorithms to control all of these variables simultaneously.
Independent variables fall into two categories, those that can be adjusted—manipulated variables (MVs)—and those that cannot—disturbance variables (DVs). Multivariable MPC uses dynamic models relating every independent variable to every dependent variable— controlled variables (CVs).
CVs also fall into two categories, control objectives and constraints. Each control objective needs to be kept as close as possible to its respective setpoint, and each constraint should approach, but not cross, the constraint value.
Multivariable MPC will reduce variability of each CV, and allow it to be held closer to the setpoint or constraint value. In those cases where there are more independent than dependent variables, extra degrees of freedom are available, and the potential for optimization exists, as many controllers can then operate in two modes: control (satisfy all constraints) or optimize (maximize an economic function and satisfy constraints).
"Yokogawa has created an alliance with Shell Global Solutions to provide a set of software packages called Exasmoc to implement multivariable MPC," says Dr. Merle Likins, PE, a solutions consultant with Yokogawa Corp. of America.
"Exasmoc interfaces Shell’s multivariable optimizing controller (SMOC) with, not only the Centum DCS, but also with any DCS that supports OPC. SMOC has been implemented in various refineries, petrochemical plants, power plants, pulp and paper plants and other facilities worldwide, with more than 1000 applications running successfully and uptime exceeding 95%."
He adds that several features distinguish it from other traditional multivariable MPC packages. "Users can combine the traditional black-box models with specific process knowledge to produce a gray-box model that will produce more accurate predictions. They can also define intermediate variables, ones that predict a change of another variable.
Likins notes that traditional multivariable MPC is designed for linear processes, whereas SMOC can handle nonlinear processes by using the Robust Quality Estimator, which not only acts as a soft sensor, but also handles nonlinear models and calculates nonlinear gains.
"Multiple economic functions may be defined. Only one economic function can be active at a time, but the economic functions may be selected, and the economic coefficient may be changed on-line. Operators can make these changes on the fly without requiring the control engineer to make the changes and download a new controller," notes Likins.
In one recent refining application, the SMOC controller was applied to a distillation unit for diesel and gas products. One controller with three sub-controllers was deployed. "An analysis was conducted to monitor operation for one month before and after commissioning. The measured economic benefit was approximately $1 million per year," reports Likins.
In another application, the SMOC controller was applied to a unit producing mono-, di- and tri-ethylene glycols. Two sub-controllers were deployed, one on the ethylene oxide side and the other on the ethylene glycol side. The measured economic benefit was approximately $3.5 million per year, primarily achieved by reducing steam consumption and increasing product yield.
"These examples demonstrate that application of multivariable MPC can produce significant benefits, with payback typically in the range of six to 12 months," concludes Likins.