A Nonconvex Regularization Scheme for the Stochastic Dual Dynamic Programming Algorithm

Abstract

We propose a new nonconvex regularization scheme to improve the performance of the stochastic dual dynamic programming (SDDP) algorithm for solving large-scale multistage stochastic programs. Specifically, we use a class of nonconvex regularization functions, namely folded concave penalty functions, to improve solution quality and the convergence rate of the SDDP procedure. We develop a strategy based on mixed-integer programming to guarantee global optimality of the nonconvex regularization problem. Moreover, we establish provable convergence guarantees for our customized SDDP algorithm. The benefits of our regularization scheme are demonstrated by solving large-scale instances of two multistage stochastic optimization problems. History: Pascal Van Hentenryck, Area Editor for Computational Modeling: Methods & Analysis. Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplemental Information ( https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2021.0255 ) as well as from the IJOC GitHub software repository ( https://github.com/INFORMSJoC/2021.0255 ). The complete IJOC Software and Data Repository is available at https://informsjoc.github.io/ .