Add an option to use Column Generation (CG) and Dual Stabilized Column Generation (DCG) as hub algorithms in mpi-sppy.
- Column Generation (CG) Hub: Implements Dantzig–Wolfe decomposition for scenario-based models. The hub builds a restricted master problem (RMP) from a subset of columns, solves for dual prices, broadcasts duals to scenario strata, and generates new columns by solving scenario subproblems. Columns are added iteratively until convergence.
References:
- A. Flores-Quiroz and K. Strunz, "A distributed computing framework for multi-stage stochastic planning of renewable power systems with energy storage as flexibility option," Applied Energy, 2021.
- K. J. Singh, A. B. Philpott and R. K. Wood, "Dantzig-Wolfe Decomposition for Solving Multistage Stochastic Capacity-Planning Problems," Operations Research, vol. 57, no. 5, pp. 1271-1286, 2009.
Dual Stabilized Column Generation (DCG) Hub: Adds a stabilized variant by solving the dual of the master problem (dRMP), with a quadratic regularization term around a reference dual value (“bundle center”). The bundle center is updated when the lower bound improves, which helps reduce oscillations in dual variables and speeds up convergence.
Reference:
- T. Schulze, A. Grothey and K. McKinnon, "A stabilised scenario decomposition algorithm applied to stochastic unit commitment problems," European Journal of Operational Research, 2017.
Pull request forthcoming with the implementation, pending examples.
Add an option to use Column Generation (CG) and Dual Stabilized Column Generation (DCG) as hub algorithms in mpi-sppy.
References:
Dual Stabilized Column Generation (DCG) Hub: Adds a stabilized variant by solving the dual of the master problem (dRMP), with a quadratic regularization term around a reference dual value (“bundle center”). The bundle center is updated when the lower bound improves, which helps reduce oscillations in dual variables and speeds up convergence.
Reference:
Pull request forthcoming with the implementation, pending examples.