A competitive learning-based Grey wolf Optimizer for engineering problems and its application to multi-layer...

Area of Science:

• Optimization Algorithms
• Artificial Intelligence
• Machine Learning

Background:

• The Grey Wolf Optimizer (GWO) is a popular meta-heuristic algorithm inspired by wolf pack behavior.
• Existing GWO variants often struggle to balance exploration and exploitation, limiting performance on complex problems.
• Population diversity is crucial for meta-heuristics to avoid premature convergence and enhance global search capabilities.

Purpose of the Study:

• To introduce a novel Competitive Learning-based Grey Wolf Optimizer (Clb-GWO) to improve the exploration-exploitation trade-off.
• To enhance population diversity and search efficiency through competitive learning strategies and a dual search system.
• To validate the effectiveness of Clb-GWO on standard benchmarking functions and real-world machine learning tasks.

Main Methods:

• Developed Clb-GWO by integrating competitive learning strategies into the standard GWO framework.
• Incorporated population sub-division into majority and minority groups with a selective complementary dual search system.
• Utilized difference vectors to promote population diversity and a better balance between exploration and exploitation.

Main Results:

• Clb-GWO demonstrated superior performance on CEC2020 and CEC2019 benchmarking suites compared to standard GWO and its variants.
• The algorithm achieved excellent results in training Multi-Layer Perceptrons (MLPs) on classification and function approximation datasets.
• Clb-GWO exhibited statistically significant improvements, lower error rates, and reduced standard deviation compared to competing methods.

Conclusions:

• The proposed Clb-GWO effectively balances exploration and exploitation, leading to enhanced optimization performance.
• The competitive learning approach and population sub-division significantly improve population diversity and search capabilities.
• Clb-GWO proves to be a robust and highly competitive meta-heuristic for both theoretical benchmarking and practical machine learning applications.