CVaR-Based Approximations of Wasserstein Distributionally Robust Chance Constraints with Application to Process Scheduling

Botong Liu,Qi Zhang,Xiaolong Ge,Zhihong Yuan

DOI: https://doi.org/10.1021/acs.iecr.0c00268

2020-04-29

Abstract:Distributionally robust chance constrained programming is a stochastic optimization approach that considers uncertainty in model parameters as well as uncertainty in the underlying probability distribution. It ensures a specified probability of constraint satisfaction for any probability distribution from a defined ambiguity set. In this work, we consider Wasserstein ambiguity sets and derive tractable approximations of individual and joint distributionally robust chance constraints (DRCCs) based on the Worst-Case Conditional Value-at-Risk (WC-CVaR). We further prove that the proposed approximations are exact for individual DRCCs and for joint DRCCs if the scaling factors are simultaneously optimized. The effectiveness of the proposed approach is demonstrated in an illustrative example and in two comprehensive process scheduling cases. The computational results show that the WC-CVaR approximation exhibits significantly better out-of-sample performance than the commonly used Bonferroni approximation, which tends to be overconservative, and classical chance constrained programming.

engineering, chemical

What problem does this paper attempt to address?

This paper aims to solve the problem of how to deal with uncertainty in chemical production scheduling. Specifically, the author proposes a method based on Conditional Value - at - Risk (CVaR) to approximate Wasserstein Distributionally Robust Chance Constraints (DRCCs) in order to deal with uncertain factors in process scheduling. The main contributions of the paper are as follows:

1. **Propose CVaR approximation methods for individual and joint Wasserstein distributionally robust chance constraints**: By using Worst - Case CVaR, the author derives tractable approximate forms of these constraints. This method is applicable not only to individual constraints but also to joint constraints, and for individual DRCCs, the approximation is exact; for joint DRCCs, it is also exact when optimizing the scaling factor.

2. **Prove the effectiveness of the CVaR approximation**: Through examples and two comprehensive process - scheduling cases, the author shows that the proposed CVaR approximation method significantly outperforms the commonly used Bonferroni approximation method and the classical chance - constrained programming method in out - of - sample performance. The Bonferroni approximation method is often too conservative, especially when dealing with large - scale joint chance constraints.

3. **Solve the limitations of traditional methods**: Traditional methods for dealing with chance constraints have problems such as high computational complexity, difficulty in handling multivariate probability integrals, and non - convexity of probability constraints. By introducing the ambiguity set defined by the Wasserstein distance, the author's method can better handle these challenges, especially in cases of insufficient data or unknown distributions.

### Overview of the paper structure

- **Part 1: Introduction**: Introduces the basic concepts of Chance Constrained Programming (CCP) and its applications in multiple fields, discusses the general challenges in dealing with chance - constrained optimization problems, and leads to the concept of Distributionally Robust Optimization (DRO).

- **Part 2: CVaR approximation of individual distributionally robust chance constraints**: Derives in detail the approximation method of individual DRCCs based on Worst - Case CVaR and proves its exactness.

- **Part 3: CVaR approximation of joint distributionally robust chance constraints**: Proposes the CVaR approximation method of joint DRCCs and discusses how to optimize the scaling factor to improve the quality of the approximation.

- **Part 4: Case study**: Verifies the effectiveness and superiority of the proposed method through specific examples and chemical production scheduling problems.

- **Part 5: Conclusion**: Summarizes the main findings and contributions of the paper and points out future research directions.

### Formula summary

- **Wasserstein distance**:
\[
W(\mathbb{P}, \mathbb{Q})=\inf_{\gamma \in \Pi(\mathbb{P}, \mathbb{Q})} \mathbb{E}_{(\xi, \zeta) \sim \gamma}[\|\xi - \zeta\|_2]
\]
where $\mathbb{P}$ and $\mathbb{Q}$ are two probability distributions, and $\Pi(\mathbb{P}, \mathbb{Q})$ is the set of all joint probability distributions with $\mathbb{P}$ and $\mathbb{Q}$ as marginal distributions.

- **Worst - case CVaR**:
\[
\text{CVaR}_\epsilon(h(\xi, x))=\inf_{\beta \in \mathbb{R}}\left\{\beta+\frac{1}{1 - \epsilon} \mathbb{E}_{\xi \sim \mathbb{P}}[(h(\xi, x)-\beta)^+]\right\}