Robust Optimization with Moment-Dispersion Ambiguity
提出矩-离差模糊集，通过分离随机变量的位置、离差和支持来增强传统模型，并引入离差特征函数和独立倾向超参数，在数据有限时比经典矩集和Wasserstein集更优。
Advancing Robust Optimization with Moment-Dispersion Framework In “Robust Optimization with Moment-Dispersion Ambiguity,” Chen, Fu, Si, Sim, and Xiong present a groundbreaking approach to robust optimization by introducing the moment-dispersion ambiguity set. This framework enhances traditional models by separately defining a random variable’s location, dispersion, and support, thereby increasing flexibility in modeling uncertainty. The authors also propose a dispersion characteristic function to capture complex properties, such as sub-Gaussian and asymmetric behaviors, alongside an independence propensity hyperparameter that supports the creation of joint ambiguity sets for multiple random variables. This innovation allows for characterizing varying levels of interdependence without requiring a correlation matrix, making the model highly applicable to real-world scenarios. Numerical experiments demonstrate that their model yields less conservative decisions compared with classic moment-based sets and offers superior robustness in data-limited scenarios when contrasted with Wasserstein ambiguity sets. This work marks a major step forward in practical risk assessment and optimization under uncertainty.