Leveraging Decision Diagrams to Solve Two-stage Stochastic Programs with Binary Recourse and Logical Linking Constraints

TL;DR

This paper introduces a generalized decision diagram approach for solving complex two-stage stochastic programs with binary recourse and logical constraints, incorporating risk measures and demonstrating effectiveness through numerical experiments.

Contribution

It extends existing BDD-based methods to handle logical linking constraints and risk measures in two-stage stochastic programs with binary recourse, a novel advancement.

Findings

Effective solution of complex stochastic programs demonstrated

Incorporation of risk measures improves decision robustness

Numerical results validate the proposed methods' efficiency

Abstract

Two-stage stochastic programs with binary recourse are challenging to solve and efficient solution methods for such problems have been limited. In this work, we generalize an existing binary decision diagram-based (BDD-based) approach of Lozano and Smith (Math. Program., 2018) to solve a special class of two-stage stochastic programs with binary recourse. In this setting, the first-stage decisions impact the second-stage constraints. Our modified problem extends the second-stage problem to a more general setting where logical expressions of the first-stage solutions enforce constraints in the second stage. We also propose a complementary problem and solution method which can be used for many of the same applications. In the complementary problem we have second-stage costs impacted by expressions of the first-stage decisions. In both settings, we convexify the second-stage problems using BDDs and parametrize either the arc costs or capacities of these BDDs with first-stage solutions depending on the problem. We further extend this work by incorporating conditional value-at-risk and we propose, to our knowledge, the first decomposition method for two-stage stochastic programs with binary recourse and a risk measure. We apply these methods to a novel stochastic dominating set problem and present numerical results to demonstrate the effectiveness of the proposed methods.