Genetic Fuzzy Modelling of Li-Ion Batteries Through a Combination of Theta-DEA and Knowledge-Based Preference Ordering

Learning semi-physical fuzzy models of rechargeable Li-Ion batteries from data involves solving a complex multicriteria optimization task where the accuracies of the approximations of the different observable variables are balanced. The fitness function of this problem depends on the recursive evaluation of a set of differential equations, where fuzzy rule-based systems are embedded as nonlinear blocks. Evaluating this function is a time consuming process, thus algorithms that efficiently promote diversity and hence demand a low number of evaluations of the fitness function are preferred. In this paper, a comparison is carried out between some recent genetic algorithms, whose performances are assessed in this particular modelling problem. It is concluded that the combination of the recent $ \theta $-Dominance Evolutionary Algorithm ($ \theta $-DEA) with a Knowledge-based precedence operator, that improves the selection, is a sensible choice. Dominance relations between the Pareto fronts are assessed in terms of binary additive $ \epsilon $ -quality indicators.