Journal international d'intelligence d'essaim et de calcul évolutif

Abstrait

Multi-layer Explosion Based Fireworks Algorithm

Jun Yu* , Hideyuki Takagi and Ying Tan

We propose a new multi-layer explosion strategy inspired by various explosion patterns of real reworks to accelerate reworks algorithm (FWA). Each rework individual conducts multiple explosions to explore a local fitness landscape carefully instead of a single layer explosion used in canonical FWA. In the proposal, each rework individual generates a small number of sparks in the first layer randomly, then the generated sparks conduct the second layer explosions to generate new diverse sparks. These new sparks repeat the above operations until the number of this iteration reaches the predefined maximum layer number. Theoretically, the number of explosion layers can be set to any positive integer, and the proposed strategy expects to generate various potential sparks using the multi-layer explosion strategy without changing the total number of generated sparks. The proposed strategy can combine with not only basic FWA but also other versions of FWA algorithms easily and replace their corresponding explosion operations to develop a new version, multi-layer explosion-based FWA. To evaluate the performance of our proposal, we select a more powerful variant of FWA, Enhanced FWA (EFWA) as the baseline algorithm and combine with our proposed explosion strategy. We run our proposal on 28 benchmark functions from CEC2013 test suites of 2-dimensions (2-D), 10-D and 30-D with 30 trial runs and compare with several state-of-theart EC algorithms. The experimental results confirm that the proposed strategy is effective and promising, which can obtain a better performance for FWA in terms of convergence speed and convergence accuracy. We finally analyze composition as well as feasibility of proposal and list some open topics.