An Improved Firefly Algorithm Based on Nonlinear Time-varying Step-size

Abstract

Firefly algorithm (FA) is a novel population-based stochastic optimization algorithm and has been shown to yield good performance for solving varieties of optimization problems. Meanwhile, it sustains premature convergence because it is easily to fall into the local optima which may generate a low accuracy of solution or even fail. To overcome this defect, a nonlinear time-varying step strategy for firefly algorithm (NTSFA) is presented. It uses a nonlinear decreasing and time-varying step-size for fireflies to better balance the algorithm's ability of exploration and exploitation. Numerical simulation on 20 test benchmark functions display that the proposed algorithm can increase the accuracy of the original FA. Finally, we apply NTSFA to integrate into k-means clustering for mouse dataset. The results show that NTSFA is an effective optimization algorithm.