Evolutionary multi-objective algorithms have been widely shown to be successful when utilized for a variety of stochastic combinatorial optimization problems. Chance constrained optimization plays an important role in complex real-world scenarios, as it allows decision makers to take into account the uncertainty of the environment. We consider a version of the knapsack problem with stochastic profits to guarantee a certain level of confidence in the profit of the solutions. We introduce the multi-objective formulations of the profit chance constrained knapsack problem and design three bi-objective fitness evaluation methods that work independently of the specific confidence level required. We evaluate our approaches using well-known multi-objective evolutionary algorithms GSEMO and NSGA-II. In addition, we introduce a filtering method for GSEMO that improves the quality of the final population by periodically removing certain solutions from the interim populations based on their confidence level. We show the effectiveness of our approaches on several benchmarks for both settings where the knapsack items have fixed uniform uncertainties and uncertainties that are positively correlated with the expected profit of an item.
翻译:当用于各种随机组合优化问题时,广泛证明进化多客观算法是成功的。机会限制优化在复杂的现实世界情景中发挥着重要作用,因为它使决策者能够考虑到环境的不确定性。我们考虑一种具有随机利润的knapack问题版本,以保证对解决方案的利润具有一定的信心。我们引入了利润机会限制 knapsack问题多目标公式,并设计了三种双目标健康评价方法,这些方法独立于所需的具体信任水平。我们使用众所周知的多目标进化算法GSEMO和NSGA-II来评估我们的方法。此外,我们为GSEMO引入了一个过滤方法,通过定期从临时人群中排除基于其信任水平的某些解决方案来提高最终人口的质量。我们展示了我们对于两种环境在 knapsack项目已经固定了统一的不确定性和不确定性并且与项目预期的利润具有积极关联性的情况下,在两种环境下采用的若干基准方法的有效性。</s>