Parameter Identification for Simulation Models by Heuristic Optimization

Abstract

In this publication we describe a generic parameter identification approach that couples the heuristic optimization framework HeuristicLab with simulation models implemented in MATLAB or Scilab. This approach enables the reuse of already available optimization algorithms in HeuristicLab such as evolution strategies, gradient-based optimization algorithms, or evolutionary algorithms and simulation models implemented in the targeted simulation environment. Hence, the configuration effort is minimized and the only necessary step to perform the parameter identification is the definition of an objective function that calculates the quality of a set of parameters proposed by the optimization algorithm; this quality is here calculated by comparing originally measured values and those produced by the simulation model using the proposed parameters. The suitability of this parameter identification approach is demonstrated using an exemplary use-case, where the mass and the two friction coefficients of an electric cart system are identified by applying two different optimization algorithms, namely the Broyden-Fletcher-Goldfarb-Shanno algorithm and the covariance matrix adaption evolution strategy. Using the here described approach a multitude of optimization algorithms becomes applicable for parameter identification.