Due to the highly dependence on the hydrological regimes, the uncertainty associated with energy planning in Brazil requires stochastic modeling of associated time series appropriately and consistently. It is clear, therefore, the importance of models to generate hydrologic scenarios to be used in the optimization via Stochastic Dual Dynamic Programming (SDDP), which improves the performance of system operations, with consequent increase in benefits and reliability and, above all, cost reduction. This stochastic modeling is performed by the PAR(p), which sets an autoregressive model of order p for each of the stages of the historical series that make up the system settings. It was shown in this work that the structure used in the simulation process of synthetic series of the model prevailing in SEB via lognormal distribution generates a nonlinearity relationship in the model equation, which causes the inconvenience of nonconvexity in the calculation of Expected Cost-to-go Functions, convex polyhedral approximated by piecewise linear functions. Considering the above and the characteristics of the stochastic model that generates the scenarios tree and its use in the optimization algorithms, this study aims the development of an alternative methodology for the construction of scenarios, so that the aforementioned drawbacks were eliminated. It is proposed a new general approach for the construction of trees, considering the steps Forward and Backward, fundamental in the process of optimization technique employed by SDDP. The structure of stochastic simulation technique developed conjugates computationally intensive Bootstrap method and Monte Carlo simulation. Scenarios trees were generated consistent with the medium-term planning of hydrothermal dispatch. The synthetic series were compared to the historical data through a battery of statistical tests and the goodness fiting of the series generated was tested that confirmed the suitability of the developed model with respect to the stochastic problem. Finally, the paths of the trees were applied to the SDDP and response variables were analyzed, leading to the conclusion that the model was able to perfectly reproduce structures compatible with the current model, but without causing the aforementioned non-linearity of the PAR(p) equation and possible non convexity in the Expected Cost-to-go Functions.