In this paper, we present decomposition algorithms for the solution of large scale stochastic dynamic problems, with convergence results and implementation suggestions. These algorithms are based on projected sub-gradient and bundle methods, applied to a dual formulation.