Model-based predictive control by heuristic optimization techniques. 
Application to nonlinear and multivariable processes.
F. Xavier Blasco Ferragud
ISBN 84-699-5429-6

Summary:

The potential of the methodology of Model-Based Predictive Control (MBPC) to 
control industrial processes is becoming increasingly significant simply by 
analyzing the number and type of industrial processes that have been 
employing. This fact is a strong motivation for those researchers who wish to 
transfer their experiences to the industrial sector.

A key element and at the same time limiting factor for the MBPC methodology is 
the optimization of indices or cost functions, since as the complexity of the 
problem grows (nonlinear models, input and output restrictions, stability 
problems real time, etc..) MBPC will be required, in general, optimization 
algorithms that guarantee the global minimum in a bounded time.

This doctoral thesis is based mainly on exploring the limiting factor to 
establish new methods for MBPC-based design optimization tools that will 
address difficult issues, such as the presence of local optima in the cost 
functions.

In this context, the doctoral thesis presents the methodologies heuristics 
Simulated Annealing and Genetic Algorithms as candidates for solving such 
problems in MBPC seeking the best adaptation of these techniques to improve 
the MBPC and setting its parameters to obtain best performance with moderate 
computational cost.

The Thesis also presents certain alternative formulations of other elements of 
the methodology of MBPC, such as the formulation of cost indices and 
predictive models (including restrictions) which are addressed in a natural 
way by the optimization methods proposed in the thesis overcoming the 
constraints of traditional methods.