RSOME: Robust Stochastic Optimization Made Easy
Multi-Item Newsvendor Problem with Wasserstein Ambiguity Sets
In this example, we consider the multi-item newsvendor problem discussed in the paper Chen et al. (2020). This newsvendor problem determines the order quantity w i w_i wi of each of the I I I products under a total budget d d d. The unit selling price and ordering cost for each product item are denoted by p i p_i pi and c i c_i ci, respectively. The uncertain demand of each product item is denoted by the random variable z ~ i \tilde{z}_i z~i. Once the demand realizes, the selling quantity y i y_i yi is expressed as min { x i , z i } \min\{x_i, z_i\} min{ xi,zi}, and the newsvendor problem can be written as the following distributionally robust optimization model,
min − p ⊤ x + sup P ∈ F ( θ ) E P [ p ⊤ y ( s ~ , z ~ , u ~ ) ] s.t. y ( s , z , u ) ≥ x − z , ∀ z ∈ Z s , ∀ s ∈ [ S ] y ( s , z , u ) ≥ 0 ∀ z ∈ Z s , ∀ s ∈ [ S ] c ⊤ x = d , x ≥ 0 \begin{aligned} \min~& -\pmb{p}^{\top}\pmb{x} + \sup\limits_{\mathbb{P}\in\mathcal{F}(\theta)}\mathbb{E}_{\mathbb{P}}\left[\pmb{p}^{\top}\pmb{y}(\tilde{s}, \tilde{\pmb{z}}, \tilde{u})\right] && \\ \text{s.t.}~&\pmb{y}(s, \pmb{z}, u) \geq \pmb{x} - \pmb{z}, && \forall \pmb{z} \in \mathcal{Z}_s, ~\forall s \in [S] \\ & \pmb{y}(s, \pmb{z}, u) \geq \pmb{0} && \forall \pmb{z} \in \mathcal{Z}_s, ~\forall s \in [S]\\ & \pmb{c}^{\top}\pmb{x} = d, ~ \pmb{x} \geq \pmb{0} \end{aligned} min s.t. −ppp⊤xxx+P∈F(θ)supEP[ppp⊤yyy(