Open    Access



Assessment of Model Predictive Control Performance Criteria



Author(s)

Rafael Lopes Duarte-Barros1* and Song Won Park2

Full-Text PDF XML     637 Views

DOI:10.17265/1934-7375/2015.02.007

1. Emerson Process Management, Rio de Janeiro 20.090-003, Brazil 2. Department of Chemical Engineering, University of São Paulo, São Paulo 05424-970, Brazil

The current highly competitive environment has driven industries to operate with increasingly restricted profit margins. Thus, it is imperative to optimize production processes. Faced with this scenario, multivariable predictive control of processes has been presented as a powerful alternative to achieve these goals. Moreover, the rationale for implementation of advanced control and subsequent analysis of its post-match performance also focus on the benefits that this tool brings to the plant. It is therefore essential to establish a methodology for analysis, based on clear and measurable criteria. Currently, there are different methodologies available in the market to assist with such analysis. These tools can have a quantitative or qualitative focus. The aim of this study is to evaluate three of the best current main performance assessment technologies: Minimum Variance Control–Harris Index; Statistical Process Control (Cp and Cpk); and the Qin and Yu Index. These indexes were studied for an alumina plant controlled by three MPC (model
predictive control) algorithms (GPC (generalized predictive control), RMPCT (robust multivariable predictive control technology) and ESSMPC (extended state space model predictive controller)) with different results.

Predictive controller performance, minimum variance, capability, MPC, GPC, ESSMPC (extended state space model predictive controller).