基于按需评估的分布式昂贵约束进化优化算法数据集

昂贵优化问题是一类常见的工业优化问题，常通过代理模型辅助的演化算法(surrogate-assisted evolutionary algorithms, SAEAs)进行求解。然而，很多昂贵优化问题不仅含有昂贵目标评估，也含有昂贵约束评估，并且这些昂贵评估是分布式的。本研究定义这类昂贵优化问题为分布式昂贵约束优化问题。这类问题的分布式特征导致目标和约束的异步评估。即使部分学者开展了目标异步评估的研究，但约束异步评估的研究还没有得到关注。因此，本研究给出了分布式昂贵约束优化问题的数学定义，并提出一种基于按需评估的分布式约束优化演化算法（distributed evolutionary constrained optimization algorithm with on-demand evaluation, DEAOE），能够异步自适应的演化不同的约束。按需评估从两个方面来提高种群的收敛性和多样性。在个体选择方面，采取一种联合样本选择策略来选择潜力个体；在约束选择方面，采取一种不可行个体优先评估策略来判断哪些约束需要进一步演化。 本数据集为该算法在两个标准测试集以及四个工业问题上进行的大量实验研究得到的原始数据以及基于原始数据分析获得的分析数据，同时与七种集中式SAEAs进行对比，以表明本研究提出的DEAOE相比于集中式SAEAs有更高的求解性能和效率。Expensive optimization problems are a prevalent category of industrial optimization tasks, typically solved via surrogate-assisted evolutionary algorithms (SAEAs). However, most expensive optimization problems feature not only costly objective evaluations but also costly constraint evaluations, and all such expensive evaluations are conducted in a distributed manner. This study defines this class of problems as distributed expensive constrained optimization problems. The distributed nature of these problems gives rise to asynchronous evaluations of both objectives and constraints. While a handful of researchers have explored asynchronous objective evaluations, investigations into asynchronous constraint evaluations remain largely overlooked. Accordingly, this study first presents a mathematical formalization of distributed expensive constrained optimization problems, then proposes a distributed evolutionary constrained optimization algorithm with on-demand evaluation (DEAOE), which enables asynchronous and adaptive evolution of diverse constraints. The on-demand evaluation mechanism enhances both the convergence and diversity of the evolutionary population from two perspectives: for individual selection, a joint sample selection strategy is employed to screen promising individuals; for constraint selection, an infeasible individual priority evaluation strategy is adopted to identify which constraints require further evolutionary optimization. This dataset encompasses the raw experimental data generated by the proposed DEAOE across two standard test suites and four industrial problem sets, alongside the analytical data derived from post-processing of the raw data. Additionally, comparative experiments are conducted against seven centralized SAEAs, verifying that the proposed DEAOE outperforms conventional centralized SAEAs in terms of both solution quality and computational efficiency.

搜集汇总

背景与挑战

背景概述

该数据集针对分布式昂贵约束优化问题，提供了基于按需评估的分布式约束优化演化算法（DEAOE）的实验数据。数据包括算法在标准测试和工业问题上的原始与分析结果，并与集中式方法对比，以验证其更高的求解性能和效率。