Partially Adaptive Robust Optimization for Uncertain Fluid Processing Networks

Abstract

This paper studies the problem of controlling fluid approximations of processing networks under uncertainty. Classical robust formulations for such models rely on static (here-and-now) control decisions that must remain feasible for all realizations of uncertainty, often leading to highly conservative solutions. We propose a partially adaptive robust optimization framework that incorporates observed system information into the control policy at the planning point, enabling less conservative and more responsive decisions while retaining worst-case guarantees. Our approach begins by modeling the fluid network dynamics as a Separated Continuous Linear Program (SCLP) in the server-effort formulation. We introduce continuous-time extensions of the budgeted uncertainty set and develop three families of uncertainty structures appropriate for fluid-flow systems. For each family, we derive the corresponding robust counterpart. Since these robust counterparts include uncertainty in the objective, they cannot be solved directly by the classical SCLP--Simplex method. To address this challenge, we develop a hybrid algorithm that integrates an extended SCLP framework with the Saddle-Point Frank--Wolfe (SP--FW) method. This hybrid approach combines tractable continuous-time optimization with efficient saddle-point updates, yielding feasible and near-optimal partially adaptive robust solutions. The resulting methodology advances robust fluid control by enabling uncertainty-aware, dynamically informed decision policies while preserving computational tractability.