Abstract: It is known that many-objective optimization problems (MaOPs) often face the difficulty of maintaining good diversity and convergence in the search process due to the high-dimensional objective space. To address this issue, this article proposes a novel multiobjective framework for many-objective optimization (Mo4Ma), which transforms the many-objective space into multiobjective space. First, the many objectives are transformed into two indicative objectives of convergence and diversity. Second, a clustering-based sequential selection strategy is put forward in the transformed multiobjective space to guide the evolutionary search process. Specifically, the selection is circularly performed on the clustered subpopulations to maintain population diversity. In each round of selection, solutions with good performance in the transformed multiobjective space will be chosen to improve the overall convergence. The Mo4Ma is a generic framework that any type of evolutionary computation algorithm can incorporate compatibly. In this article, the differential evolution (DE) is adopted as the optimizer in the Mo4Ma framework, thus resulting in an Mo4Ma-DE algorithm. Experimental results show that the Mo4Ma-DE algorithm can obtain well-converged and widely distributed Pareto solutions along with the many-objective Pareto sets of the original MaOPs. Compared with seven state-of-the-art MaOP algorithms, the proposed Mo4Ma-DE algorithm shows strong competitiveness and general better performance.
The Challenge of Many-Objective Optimization
In Multiobjective Optimization (MOP), the goal is to minimize multiple conflicting objectives simultaneously. While algorithms like NSGA-II are effective for 2 or 3 objectives, they struggle with Many-Objective Optimization (MaOP) where the number of objectives ($M$) exceeds 3.
The primary hurdle is the loss of selection pressure. As the dimensionality of the objective space increases, the proportion of non-dominated solutions in a random population grows exponentially. When nearly the entire population is non-dominated, traditional Pareto-based selection cannot distinguish between solutions, leading to a random walk that fails to converge toward the true Pareto front. Furthermore, maintaining a diverse set of solutions becomes mathematically difficult as the search space volume explodes.
The Mo4Ma Framework
The proposed Multiobjective Framework for Many-Objective Optimization (Mo4Ma) addresses these challenges by collapsing high-dimensional objective spaces into a simplified two-dimensional indicator space. This transformation allows the algorithm to focus on the two most critical aspects of optimization: convergence and diversity.
1. Indicator Transformation
Instead of optimizing $M$ objectives directly, Mo4Ma calculates two "indicative objectives" for every solution:
- Convergence Indicator ($I_c$): A sum of normalized fitness values across all original objectives, representing how close a solution is to the ideal point.
- Diversity Indicator ($I_d$): Calculated using Normalized Shift-based Density Estimation (NSDE). This metric measures local crowding while accounting for objective scaling, ensuring that solutions are well-distributed even when objectives have vastly different ranges.
2. Clustering-Based Sequential Selection (CSS)
To maintain diversity and drive convergence, the framework employs a sequential selection strategy:
- Clustering: The population is partitioned into $k$ subpopulations using k-means clustering within the 2D indicator space.
- Dynamic Selection: A probability model selects clusters based on their average convergence. Within a cluster, solutions are chosen from the best non-dominated layers in the indicator space. This circular selection process prevents the algorithm from converging prematurely to a single "niche."
Experimental Evaluation
The framework was implemented using Differential Evolution (DE) as the search engine (Mo4Ma-DE) and evaluated against seven state-of-the-art algorithms. Testing utilized two rigorous benchmark suites:
- DTLZ (1–7): Featuring various Pareto front shapes including linear, concave, and disconnected.
- WFG (1–9): Complex problems involving bias, many-to-one mappings, and deceptive landscapes.
Each problem was scaled to 5, 10, and 20 objectives to test the algorithm's robustness against increasing dimensionality.
Key Results and Insights
- Superior Scalability: Mo4Ma-DE demonstrated significant competitive advantages as the number of objectives increased. On the DTLZ suite, it achieved top-tier results in 50% of the tested instances.
- Effective Dimensionality Reduction: The results validate that while $M$ objectives define the problem, a 2D meta-metric space is sufficient for effective selection logic.
- Robustness via Normalization: The inclusion of NSDE proved critical for real-world scenarios where objective scales vary wildly (e.g., one objective ranging from 0–1 and another from 0–1000), preventing the algorithm from being biased toward larger-scale objectives.
Conclusion and Future Work
Mo4Ma provides a generic, efficient ($O(MN^2)$) framework for tackling the curse of dimensionality in optimization. While the current implementation uses $k=5$ clusters, future work will explore adaptive clustering techniques to further improve performance across diverse problem landscapes. Additionally, the authors intend to integrate other optimizers, such as Particle Swarm Optimization (PSO), into the Mo4Ma architecture to evaluate its broader compatibility.
Summary updated