Multi-objective non linear dynamic model identification considering nearly optimal solutions. Application to a µ-CHP system based on PEMFC

Abstract

Abstract Solving a wide range of engineering problems can be approached from the point of view of multi-objective optimization (MOO), i.e. trying to optimize several conflicting objectives simultaneously. Solutions to these problems are not unique and the designer must choose from several optimal solutions (Pareto set), depending on his or her preferences. However, in addition to those solutions, there are almost optimal solutions that can be preferred for several reasons. For example, if the problem is multimodal, the optimization algorithm only offers one of the possible solutions. Furthermore, the problem may present a certain degree of simplification which implies that not all preferences are reflected in the minimization objectives. The nevMOGA algorithm (multiobjective genetic algorithm of the epsilon neighborhood variable) offers the possibility of finding, apart from an approximation to the Pareto optimal set, an extra set of potentially useful near optimal solutions. This result allows a final solution more closely aligned with the designer’s actual preferences. This paper shows the application of this technique to the experimental identification problem of the parameters of a complex dynamic model. In particular, it is applied to identify the thermal model of a μ-CHP (micro Combined Heat and Power) system with a PEMFC (Proton Exchange Membrane Fuel Cell) type hydrogen cell.