Application of a dual dynamic programming algorithm to staircase linear programs arising in scheduling of hydro power plants

Abstract

A dual dynamic programming (DDP) technique for staircase linear programs was applied to problems arising in multistage scheduling of hydroelectric power systems. The approach has some similarity to nested Benders' decompositions (J.F. Benders, 1962). The contribution is a computational strategy to arrange the iterations so as to improve convergence over the standard DDP approach. The application also favors use of relaxation in the subproblem, and this is discussed.<>