Effective Streamflow Process Modeling for Optimal Reservoir Operation Using Stochastic Dual Dynamic Programming

AbstractThis article presents an innovative streamflow process model for use in reservoir operational rule design in stochastic dual dynamic programming (SDDP). Model features, which can be applied independently, are (1) a multiplicative process model for the forward phase and its linearized version for the backward phase; and (2) a nonuniform time-step length that is inversely proportional to seasonal variability. The advantages are (1) guaranteeing positive streamflow values still maintaining problem linearity, and (2) making system adaptation faster during high variable periods. Model identification is straightforward, as with the additive periodic autoregressive model generally used in SDDP. The proposed model is applied on the Senegal River system for the optimal operation of Manantali Reservoir and evaluated against the streamflow process model currently used in the water management literature.