J. Castro, J. F. Monge, L. F. Escudero
A novel approach based on a specialized interior-point method (IPM) is presented for solving large-scale multistage stochastic continuous optimization problems, where uncertainty is represented in a finite set of scenarios structured in strategic multistage trees and two-stage ones rooted at the strategic nodes whose second-stage ones represent the operational scenarios. A powerful splitting variable modeling object is considered for the step variables. The core of the IPM is the solution of the Newton's direction by a mixture of Cholesky factorization and preconditioned conjugate gradient. Two types of applications are considered: supply network design and revenue management. Broad computational experience is presented, some instance’s dimensions are 880 million linear variables, 7 million quadratic variables, 22 million constraints, 3800 scenario tree nodes. Computing time for optinal solution: Proposed approach: 2.3 days requiring 167 Gb; and CPLEX v20.1: 23 days requiring 526 Gb.
Keywords: Interior Point Method, multistage multiscale stochastic continuous optimization.
Scheduled
GT11 Continuous Optimization I
June 9, 2022 10:10 AM
A12
