Population based random search techniques in process monitoring, control and optimization
Enso Ikonen,Tiivistelmä: Tutkimus pohtii partikkelifiltterien käyttöä epävarmojen dynaamisten epälineaaristen teollisuusprosessien monitoroinnissa, säädön suunnittelussa, sekä po. systeemien analysointitekniikoita.
Abstract: The research examines the use of particle filters in the monitoring and control design of uncertain dynamic nonlinear industrial processes, and the analysis of such systems.
Particle filtering (PF) techniques are based on a gridless approximation of the conditional density of the state, given the observations. These methods were first introduced by Gordon
et al. in early 1990's, and are today known by many different names, such as sequential Monte Carlo and Bayesian bootstrap filtering; in our work, we have used the term genealogical decision trees (GDT), following the vocabulary suggested by del Moral. Particle filters have a number of attractive characteristics. In particular, they are non-parametric and can represent arbitrary distributions. Particle filters are today fairly commonly used in the area of nonlinear filtering, however industrial applications are less frequent in the area of process monitoring and control. This work is related to our work on MTM's (see here).The GDT-based control approach is a random search technique for solving a sequential optimization problems. In our work, we have introduced these techniques for solving open-loop regulation and tracking control problems. The algorithm is simple to implement, the number of design parameters is limited and the values for the parameters are easy to choose. The computational load can be heavy, however, and as the approach is model based, the performance depends on the availability of a reliable state-space model.
In addition, we have suggested a GDT-based approach for regulation. The essential idea was to use GDT optimization for solving off-line a number of predictive control problems. A finite set of initial states is then constructed from these simulations, for each of which an optimal control sequence has been computed; these trajectories are then used for setting up a controller.
Based on our work, a number of GDT-based process control examples along with clear rules for tuning the algorithm parameters have been reported:
van der Vusse CSTR (open-loop and regulation control)
rapid thermal processing (RTP)
fluidized-bed combustion (multivariable control, where all states are not measurable; also with PI-SISO regulation)
two-joint robot manipulator
ABC-plant (a batch CSTR)
We are now looking at engineering applications where the particular potential of the GDT-based control/regulation can be exploited. Also applications of 'conventional' PF-techniques in state estimation in the area of process monitoring and control are of our interest. From algorithm development point of view, a potential direction is to examine the concept of neutrality in connection of the GDT approach.
Ikonen, E., E. Gómez-Ramírez and K. Najim (2009). Process regulation via genealogical decision trees. Optimal Control Applications and Methods, 30 (2), pp. 121 – 133. (published on-line 2008, DOI 10.1002/oca.848)
Najim, K., E. Ikonen and E. Gómez-Ramírez (2008). Trajectory tracking control based on a genealogical decision tree controller for robot manipulators International Journal of Innovative Computing, Information and Control, 4 (1), pp. 53-62.
Ikonen, E and J. Kovacs (2007). Learning control of fluidized bed combustion processes for power plants. In: S. Kalogirou (Ed.), Artificial Intelligence in Energy and Renewable Energy Systems, Nova Publishers, 12, pp. 395-438. (invited) ISBN: 1-60021-261-1
Najim, K., E. Ikonen and P. Del Moral (2006). Open-loop regulation and tracking control based on a genealogical decision tree. Neural Computing & Applications, 15, no. 3/4, pp. 339-349.
Ikonen, E., E. Gómez-Ramírez and K. Najim (2006). Trajectory following and regulation of chemical batch reactors via genealogical decision trees. 1st IFAC Workshop on Applications of Large Scale Industrial Systems (ALSIS’06), 30-31 August 2006, Helsinki – Stockholm, Finland – Sweden.
Ikonen, E. (2006) Particle Filtering for Open-Loop Process Control – A Users’ Guide for a Matlab-Toolbox on Genealogical Decision Tree-Based Optimization. Systems Engineering Laboratory - Report C-30. February 2006. Oulu, Finland, 56 p. ISSN 0783-5728 ISBN 951-42-8018-0
Ikonen, E., K. Najim and P. Del Moral (2005). Application of genealogical decision trees for open-loop tracking control. 16th IFAC World Congress, 4-8 July, 2005, Prague, Czech.
Ikonen, E., P. Del Moral and K. Najim (2004). A genealogical decision tree solution to optimal control problems. IFAC Workshop on Advanced Fuzzy/Neural Control (AFNC'04), 16-17 September, 2004, Oulu, Finland.
Enso 12/2009