This thesis presents a new approach for the expected-cost-to-go functions modeling used in the stochastic dynamic programming (SDP) algorithm. The proposed technique is applied to the long-term operation planning of electrical power systems. Using state space discretization, the convex hull algorithm is used for constructing a series of hyperplanes that composes a convex set. These planes represent a piecewise linear approximation for the expected-cost-to-go functions. The mean operation costs obtained by the proposed methodology for a single water inflow scenario were compared with those from the deterministic dual dynamic programming for the same inflow scenario.This sensitivity analysis shows the convergence of both methods and is used to determine the minimum discretization level necessary to model the expected-cost-to-go functions. From the obtained result the proposed methodology is extended to the analysis of a set of water inflow scenarios, where the expected-cost-to-go function is obtained by the mean operation cost to all the considered scenario in each discretization level. The applicability of the proposed methodology for two hydro plants in a cascade is demonstrated. Additionally, a case study using code parallelization is presented aiming at gaining computational performance, where the parallelization performance, as speedup and efficiency are measured. To finish with a simulation with the whole Brazilian electrical system considering aggregated reservoir is presented.