Abstract
In this work a new approach to address multivariable control structure (MCS) design for medium/large-scale processes is proposed. The classical MCS design methodologies rely on superstructure representations which define sequential and/or bilevel mixed-integer nonlinear programming (MINLP) problems. The main drawbacks of this kind of approach are the complexity of the required solution methods (stochastic/deterministic global search), the computational time, and the optimality of the solution when simplifications are made. Instead, this work shows that, by using the sum of squared deviations (SSD) as well as the net load evaluation (NLE) concepts, the control structure design problem can be formulated as a mixed-integer quadratic programming (MIQP) model with linear constraints, featuring both optimality and improved computational performance due to state-of-the-art solvers. The formulation is implemented in the GAMS environment using CPLEX as the selected solver and two typical case studies are presented...
Highlights
All industrial processes need to be controlled in some extent if some specific behavior is required for the plant as a whole
Several contributions with the most varied themes and results appear in the literature, which include: stability/controllability/robustness from the steady-state as well as dynamic point of view[3,4,5], input-output pairing problems for decentralized square multivariable control structure (MCS) designs[6,7,8,9], self-optimizing control methods focused on using all the available manipulated variables (MVs)[1,10], deviation-based indexes for arbitrary MCS11,12, combinations of the former into multi-objective criteria[1,11,13], integration between heuristics and steady-state/dynamic simulations[14,15], and economic plantwide control where the MCS design is integrated to optimal/safe process operation[16,17,18,19]
In this work a mixed-integer quadratic programming (MIQP) model for multivariable control structure (MCS) design is proposed as a simplified reformulation of a previous bilevel mixed-integer nonlinear programming (BMINLP) superstructure[12]
Summary
All industrial processes need to be controlled in some extent if some specific behavior is required for the plant as a whole. The methodologies supported by mathematical and/or systems theory are more suitable to deal with medium-large scale processes systematically, as well as to be integrated with other problems such as process synthesis/design On this broad line, several contributions with the most varied themes and results appear in the literature, which include: stability/controllability/robustness from the steady-state as well as dynamic point of view[3,4,5], input-output pairing problems for decentralized square MCS designs[6,7,8,9], self-optimizing control methods focused on using all the available manipulated variables (MVs)[1,10], deviation-based indexes for arbitrary MCS (decentralized, sparse, full)[11,12], combinations of the former into multi-objective criteria[1,11,13], integration between heuristics and steady-state/dynamic simulations[14,15], and economic plantwide control where the MCS design is integrated to optimal/safe process operation[16,17,18,19]. In this context, the main contributions of the present work are summarized
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have