Edgar C. Tamayo

Model predictive control relevant identification and validation

The role of data prefiltering in model identification and validation is presented in this paper. A model predictive control relevant data prefilter, namely the multistep ahead prediction filter for optimal predictions over every step within a finite horizon, is presented. It is shown that models that minimize the multistep prediction errors can be identified or verified by filtering the data using certain data prefilters and then applying the prediction error method to the filtered data. Based on these identification results, a predictive control relevant model validation scheme using the local approach is proposed. The developed algorithms are verified through simulations as well as industrial applications.

Model validation for industrial model predictive control systems

This paper is concerned with model validation for industrial model predictive control systems. A new detection statistic is derived for validation of the plant model regardless of how the disturbance model changes. By appropriate filtering of process data, it is shown that performance of the on-line model validation and change detection algorithm can be improved. The proposed algorithm is illustrated by simulated examples as well as applications to model validation of an industrial model predictive control system.

An investigation into the poor performance of a model predictive control system on an industrial CGO coker

Abstract In this paper, the successful trouble shooting of an industrial model predictive control system is reported. The approach is completely data driven. Routine closed-loop operating data is the only information required for applying such a diagnosis. The source of the problem has been attributed to inappropriate selection of the disturbance variables for the MPC controller. The problem is not unusual in industrial model predictive control systems. It is therefore recommended that such an analysis is carried out for other industrial MPC control systems as well if similar problems are identified.

ASSESSMENT OF ECONOMIC PERFORMANCE OF MODEL PREDICTIVE CONTROL THROUGH VARIANCE/CONSTRAINT TUNING

Abstract Multivariate controller performance assessment (MVPA) has been developed over the last several years, but its application in advanced model predictive control (MPC) has been very limited mainly due to issues associated with comparability of variance control objective and that of MPC. MPC has been proven as one of the most effective advanced process control (APC) strategies to deal with multivariable constrained control problems with an ultimate objective towards economic optimization. Any attempt to evaluate MPC performance should therefore consider constraints and economic performance. This work is to establish a link between variance control and MPC in terms of economic performance. We show that the variance based performance assessment may be transfered to economic assessment of MPC. Algorithms for economic performance assessment and tuning are developed through linear matrix inequalities using routine operating process data. The proposed algorithms are illustrated via an industrial MPC application example.

Controller Performance Analysis Technology for Industry: Implementation and Case Studies ⋆

Abstract In this paper, an industrial MPC performance monitoring technology is introduced with a focus on the industrial implementation. A plant-oriented framework for APC performance monitoring is proposed on the basis of industrial computer control systems background. A software package integrating this technology, which is called Performance Analysis Toolbox and Solutions (PATS), is introduced. The major components of PATS are discussed including process data collection, data preprocessing, process model identification, similarity clustering, control valve stiction detection, multivariate controller performance assessment, and APC economic performance assessment using linear matrix inequality optimization. An industrial case study of a hydrogen unit is illustrated. A limited trial version of the software package can be downloaded from the web http://www.ualberta.ca/~bhuang/research/research.htm

BAYESIAN APPROACH FOR CONSTRAINT ANALYSIS OF MPC AND INDUSTRIAL APPLICATION

Abstract Profit margins from plant operations may be improved by changing the constraints so as to increase the degrees of freedom for control. Due to the presence of disturbances the chances of operating the plant outside the set limits cannot be ruled out. Thus, the expected return should be estimated by taking into account the variability. Bayesian Statistics can be used to estimate these probabilities subject to changes in operating constraint limits. The maximum a posteriori estimate of the process state due to the change in the operating conditions can be inferred using Bayesian methods and the profits or return thus obtained can be estimated. Also the decisions to obtain target value of the return can be made using the Bayesian methods.