Advancements in Search Algorithms: A Comprehensive Study

Abstract

There is an increase in the use of heuristic and metaheuristic approaches to tackle dynamic optimization problems in recent years. Since the results of an optimization process highly depend on the employed methods, the choice of a suitable algorithm for a given problem has an important effect on the quality of the proposed solutions. In the light of those facts, this study is dedicated to the performance comparison of twenty-two heuristic and metaheuristic search algorithms developed for static environments in dynamic scenarios, where the target landscapes are time-varying. The effectiveness of the algorithms is evaluated in terms of the search quality, solution diversity, robustness, adaptability, and control of the algorithmic parameters, in the face of a particular type of ‘time-varying random’ perturbations.

The static nature of the widely-known Dual Annealing and Basinhopping results, dealing with the Lennard-Jones clusters, gives a firm baseline for evaluating the effectiveness of the more recent and inherently more dynamic search approaches such as the Hybrid DE Strategy and the Adaptive Metaheuristic Framework. The study points to the need for caution when generalizing any new results on different benchmark landscape classes. A challenging dynamic benchmark suite of seven different shapes and sharply contrasting topographical evolution rates, devised herein for evaluating adaptive single-trajectory search processes, sheds new light on their performance. Two different adaptive strategies - employing a straightforward and a more sophisticated module for changing defences - indicate the synergistic benefits to be had from a sophisticated real-time adaptation of an already powerful single-trajectory