A Genetic-Algorithm Based Approach For Generating Fuzzy Singleton Models MIGUEL RAMIREZ Universidad de los Andes Facultad de Ingeniera La Hechicera - Merida VENEZUELA Ramirezmb@pdvsa.com ELIEZER COLINA Universidad de los Andes Facultad de Ingeniera La Hechicera - Merida VENEZUELA Ecolina@ula.ve Abstract: Methods for generating fuzzy singleton models from input-output data have been proposed by many authors. This paper introduces a genetic algorithm (GA) based method to generate a fuzzy singleton model tak- ing into account all the necessary constraints to guarantee an analytically inverted representation of the process dynamics which may be used as a fuzzy controller in Internal Model Control (IMC) strategy. A major advantage of this sort of models is its high interpretability compared to first-order Takagi-Sugeno fuzzy models generated from fuzzy clustering techniques [15]. The proposed method is applied to a liquid level control problem in an oil production separator based upon real input-output data, where obtaining an adequate fuzzy model is of crucial importance. Key–Words: Genetic algorithm, internal model control, fuzzy inverse control 1 Introduction Inverse fuzzy model based control has been widely used in many applications and its succesful imple- mentation depends on the exactness of the fuzzy model obtained from input - output data [1][2][3][4]. As a matter of fact, the simplest way to control a process using a fuzzy model is to invert it and use it as a controller but it is imperative to have an ideal model in order to guarantee a perfect control, which is very difficult to achieve because an exact inversion of the process can only be found in special situations and the model is never identical to the process, resulting in model-plant mismatches. In general, the output of a singleton fuzzy model is linear in the consequent parameters. This fact allows a straightforward application of standard recursive least-squares algorithms for estimating the consequent parameters from data. In this sense, a for- mal methodology was proposed in [5] based on least squared techniques. Fuzzy singleton models obtained from this methodology have a good generalization characteristic and acceptable exactness while keeping a high interpretability. In order to enhance the exactness of this sort of models which facilitates calculating its inverse, the use of genetic algorithms (GA) is proposed in this work. Genetic algorithms are randomized search algorithms that are based on the mechanics of natural selection and genetics [6][7]. They combine the principles of natural selection based on the survival of the fittest with a randomized information exchange in order to form a search and optimization algorithm. A significant number of papers have proposed dif- ferent methods to build a fuzzy rule base using GA [8][9][10][11]. Although genetic algorithms can be used for a variety of purposes, their most important application is in the field of optimization, because of their ability to seek efficiently e in large no convex search spaces, which makes them more suitable with respect to more conventional optimization techniques. In fact in many engineering fields, optimization is the basic concept behind the application of genetic algorithms (GAs) or any other evolutionary algorithm. This paper is organized as follows. Problem formulation is stated in section 2, where all the constraints of fuzzy singleton models that guarantees its exact inversion are described. The generation of the singleton fuzzy model using GA will be presented in section 3. In section 4, the fuzzy singleton model and its inverse are applied in Internal Model Control strategy. Finally, conclusions are given in section 5. Advances in Computational Intelligence, Man-Machine Systems and Cybernetics ISBN: 978-960-474-257-8 177