An Improved Artificial Bee Colony Algorithm Applied to Engineering Optimization Problems

This work proposes an improved artificial bee colony (ABC) algorithm, called the rank-based ABC algorithm, which includes a rank-based selection mechanism in the onlooker bees phase and a modified abandonment mechanism in the scout bees phase for solving unconstrained and constrained optimization problems. In the onlooker bees phase, the probability that an onlooker bee selects a food source is determined using a nonlinear selective pressure function, which is based on a ranking of fitness instead of proportional total fitness values. A nectar source with a superior fitness rank has a large probability of being selected by onlooker bees as new solutions and so yields a similar “best solution pool,” which often comprises the best and several good solutions, therefore, the exploitation capability for searching good solution is enhanced for the basic ABC algorithm. Moreover, the modified abandonment mechanism is used in the scout bees phase to increase the exploration capability for searching good solution. Accordingly, this work makes the two modifications to the basic ABC algorithm, resulting in well-balanced exploitation and exploration capabilities. Nine benchmark unconstrained problems that involve low-, midand high-dimensional optimization functions, are evaluated to compare the computational performance of the proposed rank-based ABC, the basic ABC, a crossover type ABC (CABC), a new ABC (NABC), and a Gbest-guided ABC (GABC) algorithms. Furthermore, the proposed rank-based ABC that is applied to constrained optimization problems is obtained by adding a dynamic penalty function and Deb’s constraint handling rules to the unconstrained version of ABC algorithm. Five widely used engineering design problems are solved using the five aforementioned ABC algorithms. The optima of the objective function and their standard deviations that are obtained using the proposed rank-based ABC algorithm are compared with those obtained using the four ABC algorithms mentioned above and other evolutionary algorithms.