Automatic Intelligent Initialization for a Modified Generalized Minimum Variance Controller Marco Antonio Paz Ramos (1) (2) , Enrique Quintero-Mármol Márquez (2) , José Torres Jiménez (3) (1) Instituto Tecnológico y de Estudios Superiores de Monterrey campus Ciudad de México Calle del Puente 222, Ejidos de Huipulco, Tlalpan, México D.F, México. e-mail: marco.paz@itesm.mx (2) Centro Nacional de Investigación y Desarrollo Tecnológico Interior Internando Palmira S/N, 62490, Cuernavaca Morelos, México. e-mail: eqm@cenidet.edu.mx (3) Instituto Tecnológico y de Estudios Superiores de Monterrey campus Cuernavaca Paseo de la Reforma 182-A Lomas de Cuernavaca, Temixco Morelos, México. e-mail: jtj@itesm.mx Abstract When a self-tuning regulator (STR) is applied, is essential an adequate initialization. The initialization procedure must to fix the initial parameters, to select a model structure and to propose the desired control criterion. If the goal is to have adaptive control when the knowledge about the process is scarce or even do not exist, is essential to develop a tool that can initialize to the adaptive controller in autonomous way. This paper describes an initialization approach which is based in heuristic rules and intelligent procedures. Also is developed a modification to the generalized minimum variance controller with pole assignment approach to reduce the number of parameters that must be initialized and simplify the implementation for practical cases. Test results are included in the document with the aim to highlight the characteristics of the proposed initialization procedure. PRINCIPAL SYMBOLS A,B,C Polynomials of order n a , n b and n c corresponding to, respectively, system output, control input, and disturbance input; a 0 =c 0 =1. E x {·} expectation operation. φ(t) Generalized output function. P,Q,R Costing polynomials of order n P , n Q and n R acting on, respectively, system output, input and reference. E General polynomial acting on the reference. F General polynomial acting on the perturbation. G General polynomial acting on the output. H General polynomial acting on the input. T Pole assignment polynomial y(t), u(t), r(t) System output, input and set point, respectively. e(t) Uncorrelated zero-mean random sequence. k System time delay. t time measured in sample instants. 1 Introduction The self-tuning regulator (STR) has been studied intensively during almost three decades, the original work presented by Åström and Wittermark in 1973 [1] can be pointed out as the beginning. The adoption of the microprocessor in the construction of general purpose controllers at the end of the decade of 1970 strengthened the expectations of an increase of the adaptive control in industrial applications. Considering a current industrial context, Ender [2] mentions that almost 30% of the process controllers that work in the world are operated in manual mode. This operation decision is usually made by the final user when he found unsatisfactory control performances which can be caused for different reasons: flickering in valves, problems in actuators and transmitters, inadequate designs, addition of loops which was not considered in the initial development, dramatic environmental changes and parts aging. These arbitrary dynamics open an opportunity area for the adaptive control in the industrial environment, however the profile of these controllers and its complexity have a negative impact on the final user acceptation [3], since is not possible to force to an operator (which is not usually familiarized with the adaptive control) to make decisions that determine the performance of an adaptive control [4][5]. To develop an adaptive controller for a particular process is an advantageous task, because we can always to obtain a minimum information which can be used in the control 073-40-3016-9 © 2004 ASCC 1231