An Improved New Caledonian Crow Learning Algorithm for Global Function Optimization

Area of Science:

• Computational Intelligence
• Metaheuristic Optimization
• Nature-Inspired Algorithms

Background:

• The New Caledonian crow learning algorithm (NCCLA) is a metaheuristic inspired by crow tool-making behavior.
• NCCLA faces challenges with local optima, and insufficient convergence accuracy and precision.

Purpose of the Study:

• To enhance the convergence performance of the NCCLA.
• To address limitations in local optima avoidance and convergence precision.

Main Methods:

• Proposed an improved New Caledonian crow learning algorithm (INCCLA).
• Incorporated cosine similarity for parent selection to maintain population diversity.
• Introduced a hybrid mechanism of complete and incomplete learning for exploration-exploitation balance.
• Optimized juvenile and parent individual update strategies for improved convergence.

Main Results:

• INCCLA demonstrated significant advantages over the original NCCLA and four other leading metaheuristics.
• The algorithm showed improvements in convergence speed and precision.
• Enhanced stability was observed in the performance of INCCLA.

Conclusions:

• The proposed INCCLA effectively improves upon the NCCLA.
• INCCLA offers a superior approach to optimization problems, outperforming existing methods.
• The enhancements contribute to better diversity, exploration-exploitation balance, and convergence performance.