DynamicSDDiP.jl

This project implements a special version of stochastic dual dynamic integer programming (SDDiP) and is based on (an earlier version of) SDDP.jl from Oscar Dowson.

As standard SDDiP, our code is tailored to solve multistage stochastic mixed-integer linear problems (MS-MILPs). However, in addition to standard cuts from the literature like Benders cuts, strengthened Benders cuts and Lagrangian cuts, our code allows to generate deep Lagrangian cuts (obtained by a norm-normalized dual problem) and LN Lagrangian cuts (obtained by a linearly normalized (LN) dual problem). The theory behind these cuts as well as extensive computational experiments are presented in a preprint.

Note that this project also includes features like binary approximation of the state space, Lipschitz regularization and the generation of special non-convex cuts, obtained by projecting Lagrangian cuts from a lifted binary state space to the original state space (so-called cut projection closure). These topics are discussed in another preprint.

License

DynamicSDDiP.jl is licensed under the MPL 2.0 license.

Help

If you need help, please open a GitHub issue.

This page was generated using Literate.jl.