RSOME: Robust Stochastic Optimization Made Easy
Distributionally Robust Medical Appointment
In this example, we consider a medical appointment scheduling problem described in Bertsimas et al. (2019), where N N N patients arrive at their stipulated schedule and may have to wait in a queue to be served by a physician. The patients’ consultation times are uncertain and their arrival schedules are determined at the first stage, which can influence the waiting times of the patients and the overtime of the physician.
A distributionally robust optimization model presented below is used to minimize the worst-case expected total cost of patients waiting and physician overtime over a partial cross moment ambiguity set.
min sup P ∈ F E P [ ∑ j = 1 N y i ( z ~ , u ~ ) + γ y N + 1 ( z ~ , u ~ ) ] s.t. y j ( z , u ) − y j − 1 ( z , u ) + x j + 1 ≥ z j − 1 ∀ ( z , u ) ∈ Z , ∀ j ∈ { 2 , 3 , . . . , N + 1 } y ( z , u ) ≥ 0 , ∀ ( z , u ) ∈ Z x ≥ 0 , ∑ j = 1 N x j ≤ T \begin{aligned} \min~&\sup\limits_{\mathbb{P}\in\mathcal{F}}~\mathbb{E}_{\mathbb{P}}\left[\sum\limits_{j=1}^N y_i(\tilde{\pmb{z}}, \tilde{\pmb{u}}) + \gamma y_{N+1}(\tilde{\pmb{z}}, \tilde{\pmb{u}})\right] && \\ \text{s.t.}~&y_j(\pmb{z}, \pmb{u}) - y_{j-1}(\pmb{z}, \pmb{u}) + x_{j+1} \geq z_{j-1} && \forall (\pmb{z}, \pmb{u})\in \mathcal{Z}, ~\forall j \in \{2, 3, ..., N+1\} \\ &\pmb{y}(\pmb{z}, \pmb{u}) \geq \pmb{0}, && \forall (\pmb{z}, \pmb{u})\in \mathcal{Z} \\ &\pmb{x} \geq \pmb{0},~\sum\limits_{j=1}^N x_j \leq T && \\ \end{aligned} min s.t. P∈Fsup EP[j=1∑Nyi(zzz~,uuu~)+γyN+1(zzz~,uuu~)]yj(zzz,uuu)−yj−1(zzz,uuu)+xj+1≥zj−1y