A Dynamic Neighborhood Learning-Based Gravitational Search Algorithm.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Meta-Heuristic Search

Background:

• Meta-heuristic search (M-HS) algorithms require balancing exploration and exploitation for effective optimization.
• The Gravitational Search Algorithm (GSA), while effective, often emphasizes exploitation, leading to premature convergence and loss of diversity.
• The fixed nature of GSA's archive mechanism limits its adaptability to different evolutionary states.

Purpose of the Study:

• To address the premature convergence and diversity loss issues in the standard Gravitational Search Algorithm (GSA).
• To propose a novel DNL-based GSA (DNLGSA) that adaptively balances exploration and exploitation.
• To enhance the global optimization capabilities of GSA through dynamic neighborhood learning.

Main Methods:

• Introduced a dynamic neighborhood learning (DNL) strategy to replace the fixed archive in GSA.
• Incorporated local and global neighborhood topologies to improve exploration and achieve adaptive balance.
• Defined evolutionary states using limit value and population diversity convergence criteria, coupled with a mutation operator for escaping local optima.

Main Results:

• DNLGSA demonstrated competitive performance across 27 diverse benchmark problems compared to state-of-the-art M-HS algorithms.
• The DNL strategy effectively enhanced exploration and maintained adaptive balance between exploration and exploitation.
• Reduced computational cost by decreasing gravitational force calculations through the local neighborhood topology.

Conclusions:

• The proposed DNLGSA effectively overcomes the limitations of traditional GSA, offering improved optimization performance.
• Dynamic neighborhood learning provides a robust mechanism for adaptive exploration-exploitation balance in M-HS algorithms.
• DNLGSA presents a computationally efficient and high-performing alternative for complex optimization tasks.