<p>Planning the optimal long-term generation of energy for hydro-thermal systems involves multistage stochastic programs in large dimensions. In Brazil, these problems are solved by stochastic dual dynamic programming (SDDP) techniques, considering that only water inflows into reservoirs are uncertain. The increasing growth of distributed generation has made net electricity consumption volatile, especially in the long term. To model accurately the impact of this phenomenon on the system load, the generation problem must consider not only the water inflows but also the net demand as stochastic input. We introduce probabilistic constraints into the considered multistage program, studying two different formulations. The first model ensures net demand satisfaction with a given probability, separately for each submarket in the power system. The resulting optimization problem remains solvable using the SDDP approach, without resorting to discretizing the probabilistic constraint. In the second model the probabilistic constraint holds jointly for all submarkets. This accounts for correlations unseen with the first model, but requires a sample-based discretization of the joint chance constraint. The corresponding discrete scenario representation leads to a multistage stochastic program with mixed integer variables, solved by stochastic dual integer dynamic programming. Results on real data for the Brazilian power system show the impact of the increasing penetration of distributed generation in the mix, and how the proposed probabilistic constrained models lead, not only to more stable profiles of thermal generation, but also to lower risk of deficit over the planning horizon.</p>