A coordination of some basic heuristics using Q-learning for global optimization

It is known that none of the existing heuristics on single-objective continuous optimization problems (SOCOPs) claims 100% success, rather unsuccessful on a small subset of the target class. However, the union of the solvable problems of a few heuristics can cover the whole class of SOCOPs. In this study, some basic heuristics of different categories on SOCOPs-particle swarm optimization (PSO), whale optimization algorithm (WOA), and artificial bee colony (ABC) are coordinated using Q-learning to develop a heuristic on the same target class having 100% success rate. The approach consists of two phases, where in the first phase all three algorithms are executed independently for some iterations on separate initial populations and according to their performance on the target problem, a Q-table is prepared. At the end of the first phase, three populations are combined and a population of the same size is prepared by selecting the better individuals, which is taken as the population of the next phase. In the second phase, one algorithm is selected concerning their performance reflected in the Q-table and is applied to improve the population. According to its performance, the corresponding entry of the Q-table is updated. This process is continued for another number of iterations to obtain the solution with the desired accuracy. The algorithm is tested and trusted against a large set of test instances and it is noted that it provides 100% accurate (desired accuracy) results for all the considered test instances. Its superiority compared to some state-of-the-art algorithms is also established using statistical tools.