Maximum Dispersion, Maximum Concentration: Enhancing the Quality of MOP Solutions

Multi-objective optimization problems (MOPs) often require a trade-off between conflicting objectives, maximizing diversity and convergence in the objective space. This study presents an approach to improve the quality of MOP solutions by optimizing the dispersion in the decision space and the convergence in a specific region of the objective space. Our approach defines a Region of Interest (ROI) based on a cone representing the decision maker’s preferences in the objective space, while enhancing the dispersion of solutions in the decision space using a uniformity measure. Combining solution concentration in the objective space with dispersion in the decision space intensifies the search for Pareto-optimal solutions while increasing solution diversity. When combined, these characteristics improve the quality of solutions and avoid the bias caused by clustering solutions in a specific region of the decision space. Preliminary experiments suggest that this method enhances multi-objective optimization by generating solutions that effectively balance dispersion and concentration, thereby mitigating bias in the decision space.

💡 Research Summary

The paper addresses a long‑standing challenge in multi‑objective evolutionary algorithms (MOEAs): achieving both convergence and diversity in the objective space while avoiding bias in the decision space. Traditional MOEAs focus on approximating the Pareto front (PF) but often produce solutions that cluster in certain regions of the decision space, limiting the decision maker’s options. To tackle this, the authors introduce a novel algorithm called C‑DWU, which extends the Dominance‑Weighted Uniformity (DWU) method by incorporating a decision‑maker‑driven Region of Interest (ROI) defined as a preference cone C in the objective space.

The ROI is specified by an axis vector v and an opening angle θ, representing the decision maker’s relative importance of the objectives. Solutions whose objective vectors lie outside this cone are penalized in two stages: (1) during non‑dominated sorting, a penalty Pα,θ(ϕ) proportional to the angular deviation ϕ − θ raises the front level of out‑of‑cone solutions, making them less likely to be selected; (2) during the selection phase, the original dominance‑weighted uniformity function wd(x,x′) is modified to C‑wd, which subtracts a penalty Pβ,θ(ϕ) when either solution is outside the cone. This dual‑penalty mechanism forces the algorithm to concentrate solutions inside the ROI while still encouraging maximal dispersion (uniformity) among the selected solutions in the decision space.

The algorithmic flow mirrors the original DWU: random initialization, non‑dominated sorting, crossover/mutation, creation of a combined population of size 2N, and finally selection of N individuals using the C‑DWU heuristic. The only substantive changes are the two penalization steps described above.

Experimental validation is performed on three benchmark problems: two multimodal WFG instances (WFG4 and WFG9) and a unimodal DTLZ2, all formulated as bi‑objective constrained problems where the constraint enforces the ROI cone. Performance is measured using standard objective‑space metrics (IGD, Hypervolume) and decision‑space metrics (minimum pairwise distance, dispersion). Results show that C‑DWU consistently outperforms the original DWU and classic NSGA‑II in both sets of metrics. In particular, for the multimodal WFG9 problem, C‑DWU achieves tighter convergence to the PF within the ROI and a significantly larger spread of solutions in the decision space, demonstrating its ability to mitigate clustering bias. Sensitivity analysis on the penalty intensities α and β indicates that while the exact values influence the trade‑off, a broad range yields stable improvements.

In conclusion, the study presents a practical and theoretically sound approach to simultaneously maximize dispersion in the decision space and concentration in a user‑defined region of the objective space. By embedding preference information directly into both the sorting and selection mechanisms, C‑DWU offers a more balanced set of Pareto‑optimal solutions that are both relevant to the decision maker and diverse enough to support robust implementation. Future work is suggested on extending the cone‑based ROI to higher‑dimensional objective spaces, dynamic adaptation of the ROI during optimization, and integration into real‑time decision‑support platforms.