A Firefly Algorithm Based on Prediction and Hybrid Samples Learning.

The firefly algorithm is an optimization algorithm developed based on the interactive behavior of fireflies in nature. It is based on the simulation of the mutual attraction and flickering behavior between fireflies to obtain the optimal solution by optimizing the objective function in the problem. To address the slow convergence speed, high time complexity, and easy trapping in local optima of the firefly algorithm during the search process, this paper proposes a firefly algorithm based on prediction and hybrid sample learning (PHSFA). Firstly, fireflies with insufficient progress space will be eliminated, and new fireflies will replace them to search for a better optimal value of the search function. Secondly, a hybrid sample is designed to enable each generation of fireflies to learn and enhance the convergence speed of the algorithm while reducing time complexity. In addition, a new adaptive step size strategy is adopted to adapt to the proposed algorithm. To verify the performance of PHSFA, experiments are conducted on CEC2020 test functions. The experimental results show that PHSFA performs the best on most test functions and has better solution accuracy.