Mathematical modeling of dynamical systems includes the task of parameter estimation. The solution of this task is optimal setting of model parameters that minimize the discrepancy between the model simulation and the measured behavior of the observed system. Thus, the objective function for optimization includes computationally expensive numerical simulation of models. Standard approaches to surrogate-based optimization introduce a computationally efficient predictive model that approximates the true objective function and a static, predefined substitution strategy to decide when to use the surrogate and when the true objective. The paper introduces a meta-model framework with a substitution strategy that is dynamically adapted to the solution space of the given optimization problem. The meta model encapsulates the objective function, the surrogate and the model of the substitution strategy, as well as components for learning them. The framework can be seamlessly coupled with an arbitrary optimization algorithm without any modification: It replaces the objective function and autonomously decides how to evaluate a given candidate solution. The empirical test of the framework on three parameter estimation tasks shows that the meta model significantly improves the efficiency of optimization, reducing the number of the true objective evaluations up to an average of 77%.
COBISS.SI-ID: 33102631