Identificación Robusta de Sistemas no Lineales mediante Algoritmos Evolutivos
[EN] The identification process of the parameters of a nominal model and its uncertainty, when it is used for Robust Control, is known as Parametric Robust Identification (RI). A possible approach to RI, which is appropriate when noise statistical properties unknown and/or model error invalidate sta...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Fecha de publicación: | 2006 |
| País: | España |
| Institución: | Universitat Politècnica de València (UPV) |
| Repositorio: | RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia |
| Idioma: | español |
| OAI Identifier: | oai:riunet.upv.es:10251/131396 |
| Acceso en línea: | https://riunet.upv.es/handle/10251/131396 |
| Access Level: | acceso abierto |
| Palabra clave: | Algoritmos Evolutivos Algoritmos Evolutivos Multiobjetivo Algoritmos Genéticos Identificación Paramétrica Identificación robusta Invernaderos Modelado Optimización multimodal Optimización Multiobjetivom Sistemas no lineales INGENIERIA DE SISTEMAS Y AUTOMATICA |
| Sumario: | [EN] The identification process of the parameters of a nominal model and its uncertainty, when it is used for Robust Control, is known as Parametric Robust Identification (RI). A possible approach to RI, which is appropriate when noise statistical properties unknown and/or model error invalidate statistical approaches, is the deterministic one (Set Membership Estimation). This deterministic approach assumes that identification error (IE), differences between the simulated outputs of the model and the measured outputs of the process, although unknown, will be bounded. Therefore, the objective is to estimate the parameters set of a model which keeps the identification error bounded by a certain norm and bound. This set is known as the feasible parameter set (FPS). For linear in their parameters models, the FPS is, if it exists, a convex polytope. In nonlinear models, the polytope can be non-convex even disjoint. In this thesis a RI methodology, which permits to estimate any kind of FPS in nonlinear models when IE is bounded by several norms simultaneously, is presented. This methodology converts the RI problem into a multimodal optimization problem with optimal global infinities, which constitute the FPS. For its optimization a specific evolutionary algorithm e-GA has been developed, to characterize the FPS by means of a discrete set of models FPS^* adequately distributed along the FPS. The methodology comes accompanied by a procedure that makes easy the determination of bounds, associated to the norms of the IE, in order to guarantee an FPS\neq\emptyset. For that, the Pareto Front information, which is obtained by means of minimization norms of the IE in a multobjective context is used. To solve the multobjective problem an evolutionary algorithm e-MOGA has been developed. In addition, a nominal model of restricted interpolated projection which belongs to the FPS is proposed. It is optimal in both identification and estimation errors in the parameter space. The RI of three nonlinear models, with real data, is presented as application examples of the proposed methodology: a thermal process, a model which shows the blockage that produces a given drug on the ionic currents of a cardiac cell and a greenhouse climate model (temperature and humidity) with roses hydroponic crop. |
|---|