Improving of new operators for gravitational search ‎algorithm

Heuristic optimization techniques are inspired from nature and are algorithms that can ‎obtain results at the most suitable value or close to this value in a very short time ‎heuristically in problems with broad and large solution space without scanning the whole ‎solution space. The Gravitational Search Algorithm (GSA) among one of the heuristic ‎optimization techniques attracts great attention in the solution of engineering problems ‎in recent years. The agents called masses were defined in order to find the most ‎appropriate solution with the simulation of Newton's laws of gravity and motion. After ‎the examination and analysis studies on the GSA algorithm, it was determined that there ‎are drawbacks such as being stuck in a local minimum, break from group behaviour and ‎not performing a sensitive search. In this thesis study, three new operators that eliminate ‎or reduce each drawback were developed. In the first operator suggested, changes were ‎made in the velocity and consequently positions of the agents by creating chaotic shake ‎on the gravitational constant in cases of being stuck in a local minimum and searching ‎away from the appropriate global value. That the escape velocity was added in the ‎negative direction was ensured in order to increase the velocity of the agents that remain ‎outside or far from the group behaviour and reduce it within the group in the second ‎operator developed. Hence, the study on idealizing the flock and group approach was ‎performed in the search. The aim of the last developed operator is to ensure that the ‎total force and consequently velocity are low by activating the agents with the worst ‎mass when finding the total force of an agent with the best result value in the next ‎iteration. Three new operators developed were applied to 23 single and multiple modal ‎Benchmark test functions. Each function was activated 30 times, and the best, average ‎and median values were taken. The results obtained were compared with Lipschitzian ‎mathematical method and GA, PSO and GSA, among heuristic optimization algorithms. ‎The standard GSA of all three operators and convergence speed, standard deviation, ‎and graphical comparisons were included for each function. Furthermore, the ‎comparison of the averages of the data in two dependent groups of the standard GSA ‎and new operators was performed using the Wilcoxon signed-rank test. It was seen that ‎the results obtained yield better results than other methods.‎