Lizard-Moth Flame Optimization Algorithm for Large-Scale Global Optimization Problems

Abstract This paper introduces a disruptive strategy, namely a lizard-hunting approach, into the classical Moth Flame Optimization (MFO) algorithm. The conventional MFO emulates a moth's navigation pattern around artificial light at night but tends to face stagnation due to the flame's exploitative tendencies, often getting trapped in local optima, particularly in higher-dimensional problems. The research motivation stems from the need to disrupt small groups stuck at various local optima after a certain number of iterations. To address the limitations of the existing MFO, the proposed Lizard-Moth-Flame Optimization (L-MFO) algorithm is put forth. In L-MFO, moth positions are classified into outlier and non-outlier categories using a clustering method in each iteration. Following this categorization, non-outliers are divided into highly and less densely populated subgroups, with the densely crowded group considered closer to the solution. However, a distinctive aspect of the lizard's behaviour in L-MFO is its inclination towards the less crowded group, reflecting a slower update. When a moth detects a lizard within its range and at the same angle, moths within the group either flee in the opposite direction or move towards a densely crowded group. This strategic response mitigates the issue of stagnation, enhancing the algorithm's overall performance. The proposed L-MFO algorithm undergoes a comprehensive evaluation by being compared with other state-of-the-art meta-heuristic algorithms. The assessment involves testing on twenty-three CEC-2005 benchmarks across different dimensions (10, 30, 50, 100, 500, 1000, 2000, and 5000), eight engineering problems, and 36 CEC-2017 benchmark functions with 10, 50 and 100 dimensions. The robustness of the algorithm is examined through convergence and divergence analysis, Wilcoxon signed-rank test, Two-tailed Mann-Whitney U test, and boxplot analysis. The experimental and statistical results consistently demonstrate the superior performance of L-MFO over other algorithms.