Menu:
Multivariate Performance Assessment (MVPA)
The Multivariate controller Performance Assessment (MVPA) programme was developed to allow easy performance assessment of multivariate (multiple inputs, multiple outputs) control loops using MATLAB and the Filtering and Correlation (FCOR) Algorithm. MVPA can be used to analysis the performance of advance control algorithms, such as model predictive control (MPC). This programme was developed by the Computer Process Control Group at the University of Alberta.
The basis for MVPA is similar to that of the univariate controller performance assessment (UVPA) algorithm, which was first developed by Dr. Thomas J. Harris (1989). The MVPA method uses the spectral interactor and spectral factorisation to determine the minimum variance controller that is the benchmark. The programme returns the overall performance index as well as the individual performance index for each loop. If the performance index indicates good performance, further tuning or changes in the control algorithm may not necessary or helpful. On the other hand, if the performance index indicates poor performance, then further analysis is necessary to determine the root cause of the problems.
The algorithm used by programme can be stated in the following steps (Huang & Kadali, 2008):
- Closed-loop operating data without any setpoint changes and a process model are required.
- Time-series analysis is performed to obtain an appropriate, stable model for the system.
- The interactor matrix for open-loop process is obtained from the process model.
- The Markov parameters (infinite impulse response model) of the process are determined.
- The minimum variance matrix is calculated.
- The overall performance index is calculated as the ratio of the trace of minimum variance matrix to the trace of process variance matrix.
- The individual performance indices for each loop are obtained.
Manual
The manual for using the toolbox can be found here.
System Requirements
This programme has the following prerequisites:
- Windows XP or newer.
- MATLAB® 2006a or newer. There may be issues with more recent (MATLAB® 2009a or newer) version of MATLAB®. The programme has not been tested on older versions of MATLAB®.
- System Identification Toolbox from MATLAB®.
Download
The files can be downloaded from the Downloads section.
Developers
This software has been developed by the following people:
References
- Biao Huang and Ramesh Kadali (2008). “Chapter 8: Control Loop Performance Assessment: Conventional Approach” in Dynamic Modeling, Predictive Control and Performance Monitoring: A Data-driven Subspace Approach. Springer Verlag, ISBN: 978-1-84800-232-6
- Biao Huang (2006). “Towards optimal MPC performance: industrial tools for multivariate control monitoring and diagnosis”, speech for Academic Viewpoint, SICOP Round Tables on Asset Management, International Workshop on Solving Industrial Control and Optimization Problems, Cramado, Brazil, April 6th-7th, 2006. Download copy.
- Biao Huang and Sirish Shah (1999). Performance Assessment of Control Loops. Springer Verlag, London ISBN: 978-1-85233-639-4
- Biao Huang (1997). Multivariate Statistical Methods for Control Loop Performance Assessment, PhD Thesis, University of Alberta. Download copy.