Differential Evolution Algorithm With Tracking Mechanism and Backtracking Mechanism

Abstract

Differential evolution (DE) is a simple and effective evolutionary algorithm that can be used to solve various optimization problems. In general, the population of DE tends to fall into stagnation or premature convergence so that it is unable to converge to the global optimum. To solve this issue, this paper proposes a tracking mechanism (TM) to promote population convergence when the population falls into stagnation and a backtracking mechanism (BTM) to re-enhance the population diversity when the population traps into the state of premature convergence. More specifically, when the population falls into stagnation, the TM is triggered so that the individuals who fall into the stagnant situation will evolve toward the excellent individuals in the population to promote population convergence. When the population goes into the premature convergence status, the BTM is activated so that the premature individuals go back to one of the previous statuses so as to restore the population diversity. The TM and BTM work together as a general framework and they are embedded into six classic DEs and nine state-of-the-art DE variants. The experimental results on 30 CEC2014 test functions demonstrate that the TM and BTM are able to effectively overcome the issues of stagnation and premature convergence, respectively, and therefore, enhance the performance of the DE significantly. Moreover, the experimental results also verify that the TM works together with the BTM as a general framework is better than other similar general frameworks.