A Novel Approach for the Coordination of Block Decentralized Model Predictive Control

Abstract

<p>In this thesis, a novel block decentralized MPC approach is implemented in order to coordinate the control of interacting process units (blocks) in a chemical plant. The goal of this research is to develop coordinated control that enables each block to optimize its own performance by adjusting only is manipulated variables while accounting for interactions among blocks.</p> <p>A simultaneous algorithm, termed D-MPC, is proposed that replaces multiple optimizations (from several, interacting MPC controllers) with one set of equations, yielding a single-level optimization problem. Given the complexity of the resulting problem consisting of linear and complementarity equations, an efficient active set heuristic is proposed for real time computations. The approach is computationally tractable, yielding a small set of convex problems to be solved sequentially and providing reliable solutions with good dynamic performance for the cases studied.</p> <p>Integrity is important for control designs, and generally, block designs with negative and zero Block Relative Gains (BRG) have poor integrity and cannot be controlled with published approaches. In contrast, the D-MPC approach successfully provides good integrity for processes with all BRG signs while maintaining the desired autonomy of each individual block.</p> <p>The solution existence, uniqueness, and stability of the proposed controller are also discussed in order to delimit what kind of processes can be controlled using the proposed D-MPC controller. A simple D-MPC formulation is analyzed to demonstrate that specific ranges of controller tuning can lead to the loss· of nominal stability for negative BRG systems. Therefore, a step-wise D-MPC design procedure was developed that integrates a stability analysis first proposed for centralized MPC and successfully adapted for the D-MPC controller.</p> <p>The dynamic performance and integrity demonstrated in case studies with all signs of BRG and sizes from 2x2 to 4x4 demonstrate the computational tractability, good dynamic performance of D-MPC controller designs developed with the design procedure and implemented with the heuristic algorithm.</p>