TL;DR
This paper introduces MRIME-CD, a novel metaheuristic algorithm that enhances the RIME algorithm by incorporating covariance learning and diversity enhancement strategies. MRIME-CD significantly improves solution accuracy, convergence speed, and stability for numerical optimization problems, outperforming the original RIME and other state-of-the-art algorithms.
What Problem Does This Paper Address?
Metaheuristic algorithms are powerful tools for solving complex optimization problems that are common in various scientific and engineering fields. While the RIME algorithm, a recently proposed physics-based metaheuristic, shows promise with fast convergence, it suffers from a critical drawback: it tends to lose population diversity during optimization, leading to premature convergence and getting stuck in local optima. This imbalance between exploration and exploitation limits its effectiveness on complex, real-world problems. The authors aim to address these limitations by developing a modified RIME algorithm that maintains population diversity and improves its ability to escape local optima.
Methods
The proposed MRIME-CD algorithm introduces three key strategies to enhance the original RIME:
- Gaussian-based Covariance Learning Strategy (GCLS): This strategy is applied during the "soft-rime search" phase. It uses Gaussian-based covariance learning to increase population diversity and balance the over-exploitation tendency of RIME by leveraging the "bootstrapping effect" of dominant populations.
- Average Bootstrapping Strategy (ABS): Introduced in the "hard-rime puncture mechanism," this strategy guides the population's search in the early stages by considering the weighted positions of dominant populations. This aims to improve global search capabilities and prevent premature convergence towards a single optimal individual.
- Stochastic Covariance Learning-based Population Diversity Mechanism (SPDM): This mechanism addresses stagnation. It uses a novel stagnation indicator and a stochastic covariance learning strategy to update stagnant individuals, thereby improving the algorithm's ability to escape local optima.
Datasets
The MRIME-CD algorithm was evaluated on two standard benchmark test suites:
- CEC2017: This suite includes 29 functions, categorized into unimodal, multimodal, hybrid, and composite functions, with dimensions of 10, 30, 50, and 100.
- CEC2022: This suite includes 12 functions, categorized into unimodal, basic, hybrid, and composite functions, with dimensions of 10 and 20.
The paper also mentions an evaluation on 10 engineering constrained optimization problems.
Key Insights and Ideas
- Diversity is Key to Escaping Local Optima: The paper strongly emphasizes that a lack of population diversity is the primary reason RIME gets stuck in local optima. The proposed strategies directly target this by introducing mechanisms to maintain and enrich diversity throughout the optimization process.
- Balancing Exploration and Exploitation: The three proposed strategies work in concert to achieve a better balance between exploring new regions of the search space and exploiting promising areas. GCLS and ABS focus on exploration and early-stage search, while SPDM helps in escaping local optima during later stages.
- Synergistic Effect of Improvements: The authors demonstrate that combining the three proposed strategies (MRIME-CD) yields superior results compared to using any single strategy alone, indicating that these improvements are mutually reinforcing.
- Robustness Across Dimensions: MRIME-CD shows stable performance and minimal sensitivity to changes in dimensionality, highlighting its scalability.
Key Results
- Superior Performance: MRIME-CD consistently outperformed the basic RIME algorithm and other compared metaheuristics across various benchmark functions and dimensions.
- Accuracy and Stability: The proposed algorithm achieved significantly better solution accuracy and demonstrated superior stability in its performance.
- Convergence Speed: MRIME-CD exhibited faster convergence speeds compared to its counterparts, reaching optimal or near-optimal solutions in fewer function evaluations.
- Statistical Significance: Statistical tests (Friedman, Wilcoxon rank sum, Kruskal Wallis) confirmed that the performance differences between MRIME-CD and other algorithms were statistically significant. For instance, on CEC2017, MRIME-CD achieved an average rank of 1.51, while basic RIME ranked last.
- Engineering Application Potential: MRIME-CD showed promising results on constrained engineering optimization problems, consistently finding better design solutions and maintaining high accuracy.
Limitations and Future Work
- Computational Cost: The authors acknowledge that computing the covariance matrix can be time-consuming, making MRIME-CD more suitable for problems where computation time is not a critical constraint. They suggest exploring parallel computing to speed up the process.
- Parameter Sensitivity: Further investigation into the parameter sensitivity of MRIME-CD is recommended.
- Future Directions: The authors propose extending MRIME-CD to solve more complex engineering applications (e.g., structure optimization, task planning, image segmentation), developing multi-objective and binary versions, and further optimizing its computational cost and algorithmic complexity.
Summary updated