Infinite Horizon in Stochastic Dual Dynamic Programming
Shasa Foster, Ben Fulton, Tony Downward, and Andy Philpott
Department of Engineering Science, University of Auckland
This report discusses the extensions made to stochastic dual dynamic programming to produce an infinite horizon stochastic dual dynamic programming methodology useful for solving large multistage convex stochastic optimization problems. The infinite horizon methodology was implemented by extending SDDP.jl, an existing package (in the programming language Julia) using stochastic dual dynamic programming. Parallel processing and cut selection heuristics present in SDDP.jl were integrated into the infinite horizon extensions of SDDP.jl. The infinite horizon SDDP algorithm was then applied to a hydro-thermal scheduling model of the NZEM (the DOASA model) for various configurations of the NZEM.
This presentation is eligible for the ORSNZ Young Practitioners Prize.