Cooperative Optimization-Based Frequency-Domain Curve Fitting: Enhancing Accuracy and Efficiency

Abstract

The frequency-domain curve fitting of dynamic system models in modal summation form is an essential process in mechatronics control, and it is widely used in various industries. However, as the fitting problem often results in a nonconvex optimization problem, gradient-based methods, such as the nonlinear least-squares (NLSs) method, tend to become trapped in local optima owing to sensitivity to initial parameters, whereas metaheuristics-based methods, such as the genetic algorithm (GA), are hindered by extensive parameter search times. This article presents a novel curve fitting method based on a cooperative optimization approach that combines the GA and least-squares method. The proposed method efficiently and accurately identifies (quasi)global optimal parameters for nonconvex fitting problems without requiring complex initial parameter settings. Experimental results on a galvanometer scanner demonstrated that the proposed method outperforms conventional NLS-based and GA-based methods in terms of accuracy and efficiency.