Improved shuffled Frog leaping algorithm with unsupervised population partitioning strategies for complex optimization problems

Abstract

Shuffled Frog leaping algorithm (SFLA) is a multi population swarm intelligence algorithm which employs population partitioning techniques during the evolutionary stage. Methods adopted by SFLA for partitioning the population into memeplexes play a critical role in determining its ability to solve complex optimization problems. However, limited research is done in this direction. This work presents supervised machine learning based methods Spectral Partitioning (SCP), Agglomerative Partitioning (AGP) and Ward Hierarchical Partitioning (WHP) for distributing the solutions into memeplexes. The efficacy of variants of SFLA with these methods is assessed over CEC2015 Bound Constrained Single-Objective Computationally Expensive Numerical Optimisation problems. Analysis of results establishes that proposed SCP, AGP and WHP methods outperform Shuffled complex evolution (SCE) partitioning technique; Seed and distance based partitioning technique (SEED), Random partitioning (RAND) and Dynamic sub-swarm partitioning (DNS) for more than 10 functions. Time complexity of all the algorithms is comparable with each other. Statistical analysis using Wilcoxon signed rank sum test indicates that SCP, AGP and WHP perform significantly better than existing approaches for small dimensions.