Q-Learning-Based Adaptive Bacterial Foraging Optimization

Abstract

As a common biological heuristic algorithm, Bacterial Foraging Optimization (BFO) is often used to solve optimization problems. Aiming at increasing the solution accuracy and convergence performance while enhancing the capability of individual’s self-learning and exploration, a Q-Learning-Based Adaptive Bacterial Foraging Optimization (QABFO) is proposed in this paper. The basic chemotaxis, reproduction and elimination/dispersal operations in standard BFO are redesigned in a Q-learning mechanism, and the Q-table will be updated on the basis of the changed fitness values in each iteration. In chemotaxis operation, modified search direction and secondary cruising mechanism are introduced into tumbling and swimming behaviors, with the purpose of improving the search efficiency and balancing local and global search. Additionally, a generation skipping adaptive reproduction is designed to control the accuracy and convergence of QABFO. Experimental results demonstrate that compared with BFO, PSO and GA, the proposed algorithm performs better in terms of accuracy and stability on most of the test functions and can effectively improve the premature convergence problem due to the original reproduction operation in BFO.