Abstract
The Whale Optimization Algorithm (WOA) is a prominent swarm intelligence technique inspired by the bubble-net hunting behavior of humpback whales. Despite its popularity, the standard WOA often suffers from search range overshoot and premature convergence to local optima. This paper presents an enhanced version, ARSC-WOA, which integrates three novel spiral searching strategies—specifically the Archimedes spiral—and an improved sine-cosine operator governed by a non-linear convergence factor. Experimental results on 22 benchmark functions and two engineering design problems demonstrate that the proposed hybrid algorithm offers superior search accuracy, faster convergence, and robust stability in high-dimensional spaces.
Introduction
Meta-heuristic algorithms have become indispensable for solving complex, non-linear optimization problems where traditional gradient-based methods are ineffective. The Whale Optimization Algorithm (WOA) mimics the specialized hunting mechanism of whales. However, its reliance on a logarithmic spiral update can lead to agents jumping outside the feasible search space, reducing population diversity. Furthermore, the transition from global exploration to local exploitation is often abrupt, leading to suboptimal solutions.
Methodology: The ARSC-WOA Framework
The proposed ARSC-WOA addresses these limitations through three primary innovations:
1. Optimized Spiral Searching Strategies
To prevent search range overshoot, the authors evaluated three geometric routes: Archimedes, Rose, and Hypotrochoid spirals. The Archimedes spiral was selected for its isometric property, ensuring search agents maintain a more uniform and bounded distribution compared to the standard logarithmic spiral.
2. Improved Sine Cosine Operator and Merit-based Selection
The algorithm incorporates a Sine Cosine Algorithm (SCA) operator. Instead of a random choice, a merit-based strategy is employed: the algorithm calculates potential positions using both sine (for exploration) and cosine (for exploitation) functions, then selects the position with the superior fitness value. This reduces the randomness inherent in meta-heuristics and stabilizes convergence.
3. Non-linear Convergence Factor
A standard linear decrease in the convergence factor often fails to capture the nuances of the search process. ARSC-WOA introduces an exponential convergence factor ($c_1$), allowing for an extended exploration phase followed by a rapid transition to exploitation as the iteration count increases.
Experimental Evaluation and Results
The performance of ARSC-WOA was validated against a suite of 22 benchmark functions and compared with established algorithms such as the Salp Swarm Algorithm (SSA), Firefly Algorithm (FA), and the standard WOA.
- Benchmark Performance: ARSC-WOA achieved the theoretical optimal value in 50% of the test functions, demonstrating exceptional accuracy.
- Convergence Speed: Log-scale analysis revealed that the hybrid algorithm reaches optimal values significantly faster than its counterparts.
- Engineering Applications: In the Welded Beam Optimization problem, ARSC-WOA achieved a minimum cost of 1.713, outperforming the standard WOA (1.730) and the Gravitational Search Algorithm (GSA).
- Scalability: The algorithm maintained high performance even when problem complexity was scaled to 200 dimensions, avoiding the "curse of dimensionality."
Conclusion
By refining the geometric search path and introducing a merit-based selection mechanism with a non-linear convergence factor, the ARSC-WOA significantly enhances the robustness of the original Whale Optimization Algorithm. Future work will explore its application in multi-objective optimization and discrete binary problems, such as feature selection in machine learning.
Summary updated