Artificial Intelligence Review Artificial Intelligence
Incorporating Q-learning and gradient search scheme into JAYA algorithm for global optimization
Abstract
Many swarm intelligence techniques are facing rigorous challenges since they cannot exploit useful information well during the evolutionary procedure. To remedy this issue, this paper raises a reinforced JAYA algorithm (QLJAYA) that employs the Q-learning and gradient search scheme. In QLJAYA, to balance convergence and diversity, a modified search formula and gradient search scheme are adaptively selected to generate solutions under the control of Q-learning. In addition, to strengthen the rotational invariance of JAYA, the covariance matrix learning strategy is adopted to construct an eigen coordinate system for each solution. Experimental simulations on CEC2017 and CEC2019 test suites and application to identify parameters of photovoltaic systems suggest that QLJAYA can exhibit a better or at least competitive overall performance compared to several typical JAYA variants and other well-known metaheuristics. The source code of QLJAYA is publicly available at https://github.com/denglingyun123/QLJAYA .