A Hierarchical Method of Parameter Setting for Population-Based Metaheuristic Optimization Algorithms

Metaheuristic algorithms for a global optimization problem have unbound strategy parameters that affect solution accuracy and algorithm efficiency. The task of determining the optimal values of unbound parameters is called the parameter setting problem and can be solved by static parameter setting methods (performed before the algorithm run) and dynamic parameter control methods (performed during the run). The paper introduces a novel hierarchical parameter setting method for the class of population-based metaheuristic optimization algorithms. A distinctive feature of this method is the use of a hierarchical algorithm model. The lower level represents a sequential algorithm from this class, and the upper level represents an algorithm with the parallel island model. Parameter setting is performed by the hierarchical method, which combines parameter tuning for the sequential algorithm and adaptive parameter control for the parallel algorithm. Parameter control is based on vector fitness criteria which consist of a convergence rate and a solution value. An approach for estimating the convergence rate for a multistep optimization method is proposed. Experimental results for CEC benchmark problems are presented and discussed.