Accelerating Fireworks Algorithm with Dynamic Population Size Strategy
Jun Yu, Hideyuki Takagi · 2020
A dynamic population size strategy is proposed for the fireworks algorithm (FWA) to adjust the population size based to the search results of the current generation. When the currently found optimal individual is updated, a linear decreasing method is activated to maintain an efficient exploitation speed. The population size is reduced by 1 until the minimum preset population size is reached, then the population size remains unchanged. Otherwise, we randomly generate a larger population size than the initial population and expand the explosion amplitudes of all firework individuals artificially, which the expectation that we can escape current local minima. To analyze the effectiveness of the proposed strategy, we combined it with the enhanced FWA (EFWA) together, and run the EFWA and (the EFWA + our proposed strategy) on 28 CEC 2013 benchmark functions in three different dimensions. Each function is run 30 trial times independently, and the Wilcoxon signed-rank test is applied to check significant differences. The statistical results showed that the proposed dynamic population size strategy can not only achieve a faster convergence speed for the FWA but also can jump out of trapped local minima more easily to maintain a higher performance, especially for high-dimensional problems.