Jul 2026· 2026 IEEE 27th China Conference on System Simulation Technology and its Applications (CCSSTA)· pp. 835-840· 0 citations· 11 references
Abstract
In evolutionary algorithms (EAs), static operator configurations often depend on human expertise and may lead to search stagnation. Existing adaptive methods either overlook real-time optimization states or require substantial computational overhead. Recent LLM-based methods support adaptive decision-making; however, applying an LLM to a single population requires the model to resolve conflicting exploration and exploitation objectives, which may reduce the reliability of operator selection. To address this limitation, this paper proposes DEAL, a Dual-Pool Evolutionary Algorithm with LLM-Driven Adaptive Operator Selection. DEAL separates the population into an exploitation pool and an exploration pool, enabling the LLM to pursue different search objectives: refining promising solutions in the exploitation pool and exploring alternative candidate structures in the exploration pool. DEAL further incorporates an offline experience collection module and a structured prompt template that jointly provide operator information, population status, and relevant historical evidence. Benchmark experiments on EUC-2D TSP instances show that DEAL reduces the average optimality gap on most tested categories compared with conventional baselines, suggesting the effectiveness of dual-pool LLM-driven operator scheduling.
ADPSO-ERLS is a discrete swarm algorithm that treats this allocation as an explicit, tunable design variable, and ranks first under the Friedman test, and all twenty-five multiplicity-controlled Wilcoxon comparisons favor it with large, near-complete distributional separation.
A. Soria-Lorente, Jean-Marie Vilaire, Junior Michel et al.· 0 citations
Particle swarm optimization (PSO) tends to suffer from premature convergence and limited adaptability when addressing high-dimensional and complex multimodal optimization problems. To alleviate these issues, a hybrid framework termed MDLPSO-DE is proposed by integrating memory-driven distribution learning with differential evolution and adaptive operator selection. An archive mechanism is employed to retain high-quality historical solutions and provide stable evolutionary guidance. A rank-weighted distribution model constructed from elite individuals is used to enhance exploitation while maintaining population diversity. A differential evolution mutation operator is incorporated into the velocity–position update process to improve global search capability. Furthermore, operator activation probabilities are dynamically adjusted according to their historical contribution, enabling adaptive coordination between exploration and exploitation during different search stages. The proposed method is evaluated on the CEC2017 benchmark suite under multiple dimensional settings. Comparative results indicate that MDLPSO-DE achieves improved optimization accuracy and convergence behavior relative to several representative swarm and evolutionary algorithms. Statistical analysis based on the Friedman test and Holm-corrected Wilcoxon signed-rank tests confirms that the performance advantage of MDLPSO-DE is statistically significant across the majority of benchmark functions and baseline algorithms. Ablation experiments are conducted to examine the contribution of each module. An application-inspired learning path optimization case study is further used to demonstrate the applicability of the proposed method under structured and constrained optimization settings.
Lanhua Ding, Weitao Zhao· Journal of King Saud Univers...· 0 citations
This work investigates how balancing the number of decomposed subproblems and ants impacts performance under a fixed computational budget and introduces MOEA/D-AM, a hybrid algorithm that adapts this decomposition strategy for partial classification rule induction.
Constrained multimodal multiobjective optimization problems (CM-MOPs) widely exist in real-world applications and are characterized by the coexistence of constraints and multimodality. Solving CMMOPs requires identifying multiple feasible Pareto-optimal solutions with identical objective values. However, many existing algorithms tend to converge prematurely to local feasible regions and fail to discover all equivalent Pareto-optimal solutions. To address this issue, this paper proposes a dynamic grid search-based evolutionary algorithm (DGSEA) for CMMOPs. DGSEA assigns a dynamic grid space to each solution, which expands as the evolution progresses. In the early stage, a small grid promotes effective exploration of discrete feasible regions while maintaining a well-distributed set of candidate solutions. In the middle and later stages, the expanded grid helps eliminate redundant solutions and achieves a better balance among feasibility, convergence, and diversity. Moreover, a grid-based density metric is incorporated into mating and environmental selection to generate and select offspring with good distribution. Experimental comparisons with state-of-the-art algorithms demonstrate that DGSEA achieves superior performance in solving CMMOPs.
J. Zou, Yu Li, Hui Bai et al.· GECCO Companion· 0 citations
Constrained optimization problems (COPs) are widely encountered in real-world applications and remain challenging to solve. Accordingly, developing efficient solution methods for COPs continues to be an important research topic. This paper proposes a novel hybrid DE-PSO algorithm, termed Dual-Population DEPSO (DP-DEPSO), to address COPs. DP-DEPSO is built upon two key ideas. First, a dual-population scheme is adopted in which Differential Evolution (DE) and Particle Swarm Optimization (PSO) evolve in parallel. Second, population-distinct constraint-handling schemes are introduced, where the DE group employs Deb's feasibility rule to preserve and explore high-quality feasible solutions, while the PSO group utilizes Deb's rule and an ε constraint handling method to exploit promising regions. Through complementary search behaviors of the dual population, DP-DEPSO achieves a well-balanced exploration-exploitation search. The proposed method is evaluated on thirteen classical constrained benchmark problems and compared with state-of-the-art evolutionary algorithms and existing DEPSO. Experimental results demonstrate that DP-DEPSO attains superior solution quality and competitive convergence efficiency. Statistical analyses using the Wilcoxon signed-rank test further confirm its significant performance advantage over most comparison methods. These results indicate that DP-DEPSO provides an effective and robust framework for COPs.
So Fukuhara, Masao Arakawa· GECCO Companion· 0 citations
Heuristic design for combinatorial optimization remains heavily reliant on expert knowledge, while existing large language model (LLM)-enhanced evolutionary methods typically evolve isolated algorithmic components, even when one determines the search state on which another operates. This paper proposes LLM-driven Heuristic Components Joint Generation (LLM-HCJG), a population-based framework that jointly generates and co-evolves interdependent heuristic components under a shared design blueprint. Applied to guided local search (GLS), LLM-HCJG couples solution initialization with penalty construction and embeds the generated pair into an enhanced online search mechanism. The resulting form is further transferred from the traveling salesman problem (TSP) to the capacitated vehicle routing problem (CVRP). Theoretical analysis establishes the non-separable state-transition effects between the two components and the advantage in generation consistency. Across synthetic instances and 41 public TSPLIB/CVRPLIB benchmarks, LLM-HCJG attains consistently low optimality gaps, including best or tied-best results on 28 of 29 TSPLIB instances and all 12 CVRPLIB instances. Ablation and structural analyses further indicate that these gains are associated with cross-component compatibility and alignment rather than isolated-component recombination. These results support effective cross-instance transfer within the evaluated routing settings under limited-sample, modest-cost training.
Junyi Wei, Yangming Zhou, Zhi-Bin Jiang et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
