Oversizing analysis in plant-wide control design for industrial processes

Abstract

In this work, an alternative plant-wide control design approach based on oversizing analysis is presented. The overall strategy can be divided in two main sequential tasks: 1 – defining the optimal decentralized control structure, and 2 – setting the controller interaction degree and its implementation. Both problems represent combinatorial optimizations based on multi-objective functional costs and were solved efficiently by genetic algorithms. The first task defines the optimal selection of controlled and manipulated variables simultaneously, the input–output pairing, and the overall controller dimension in a sum of square deviations context. The second task analyzes the potential improvements by defining the controller interaction degree via the net load evaluation approach. In addition, some insights are given about the feasibility (implementation load) of these control structures for a decentralized or centralized framework. The well-known Tennessee Eastman (TE) process is selected here for sake of comparison with other multivariable control designs.