TL;DR
This paper introduces MSIGOA, a novel metaheuristic algorithm designed to overcome the limitations of the original Gazelle Optimization Algorithm (GOA). By incorporating an iteration-based updating framework, adaptive parameter tuning, and a dominant population-based restart strategy, MSIGOA achieves a better balance between exploration and exploitation, leading to improved convergence speed and accuracy on complex optimization problems.
What Problem Does This Paper Address?
Many real-world problems, especially in engineering, involve complex optimization challenges characterized by nonlinearity, nonconvexity, and multiple constraints. Traditional optimization methods often struggle with these complexities, getting stuck in local optima or requiring specific function properties (like differentiability) that are not always present. Metaheuristic algorithms, inspired by natural phenomena, offer a more flexible approach. The Gazelle Optimization Algorithm (GOA) is a recent metaheuristic that shows promise, but it suffers from an imbalance between exploration (searching broadly) and exploitation (refining solutions), random switching between these phases, and insufficient information exchange within the population, which can lead to premature convergence and getting trapped in local optima. This paper aims to address these shortcomings by proposing a multi-strategy improved GOA.
Methods
The authors introduce three key enhancements to the original GOA to create MSIGOA:
- Iteration-based Updating Framework (IBUF): Instead of randomly switching between exploration and exploitation, MSIGOA divides the optimization process into three stages. The early stage focuses on global exploration, the middle stage balances exploration and exploitation, and the final stage emphasizes local exploitation for fine-tuning solutions.
- Adaptive Parameter Tuning Strategy (APTS): Two adaptive strategies are employed to dynamically adjust parameters for Brownian motion and Lévy flight. This allows the algorithm to better explore the search space and avoid drastic step-size changes, leading to a smoother optimization process and improved adaptability.
- Dominant Population-based Restart Mechanism (DPRM): To combat premature convergence and the tendency to get stuck in local optima, MSIGOA incorporates a restart mechanism. This strategy leverages the optimal agent, the weighted position of the dominant population (top 50% of agents), and the agent itself to guide the search towards more promising regions and maintain population diversity.
Datasets
The paper evaluates MSIGOA on two standard benchmark test suites:
- CEC2017: A comprehensive set of 41 test functions covering unimodal, multimodal, hybrid, and composition categories.
- CEC2022: A set of 12 benchmark test functions.
Additionally, MSIGOA's performance is tested on three engineering design optimization problems:
- Welded beam design problem
- Pressure vessel design problem
- Tension/compression spring design problem
Key Insights and Ideas
- Structured search is better than random: The IBUF demonstrates that a structured, stage-based approach to balancing exploration and exploitation is more effective than random switching, leading to better overall performance.
- Adaptability is crucial: Dynamically adjusting algorithm parameters (APTS) significantly improves its ability to handle diverse optimization landscapes and avoid getting stuck.
- Leveraging collective intelligence: The DPRM highlights the power of using information from the best-performing individuals and the dominant population to guide the search and escape local optima.
- Dimensionality matters: The effectiveness of the DPRM is shown to be influenced by the problem's dimensionality, with higher dimensions often benefiting more from a larger dominant population.
Key Results
MSIGOA consistently outperforms the basic GOA and several other advanced metaheuristic algorithms across the tested benchmark functions and engineering problems.
- On CEC2017 and CEC2022 test sets, MSIGOA was not worse than GOA in 92.2% and 83.3% of functions, respectively.
- MSIGOA was not worse than other advanced algorithms in 88.57% and 87.5% of functions on CEC2017 and CEC2022, respectively.
- Statistical tests (Friedman and Wilcoxon rank-sum) confirm MSIGOA's superior performance and stability, especially in higher dimensions.
- In engineering design problems, MSIGOA achieved optimal or near-optimal solutions, demonstrating its practical applicability.
Limitations and Future Work
The authors acknowledge that MSIGOA, while improved, still faces challenges:
- It may not provide the absolute best or most stable solutions for all functions.
- It can consume more time solving complex optimization problems compared to simpler algorithms.
Future work suggested includes:
- Exploring MSIGOA's potential in multi-objective optimization scenarios.
- Developing hybrid strategies that integrate MSIGOA with other advanced metaheuristics or machine learning methods.
- Further analysis using more comprehensive experimental setups, such as the GKLS function generator.
Summary updated