Data-Driven Multistage Distribuionally Robust Programming to Hydrothermal Economic Dispatch With Renewable Energy Sources

The multistage solution is very important to achieve optimal hydrothermal economic dispatch considering the uncertainty of renewable energy sources. In data-driven settings, only some historical trajectories are available and the probability distribution is unknown. A data-driven scheme for multistage stochastic hydrothermal economic dispatch with Markovian uncertainties is proposed in this paper. Then a data-driven distributionally robust stochastic dual dynamic programming (DDR-SDDP) is proposed to tackle the corresponding computational intractability, where the conditional probability distributions are estimated by using kernel regression. The out-of-sample performances are improved by distributionally robust optimization on a Wasserstein distance-based ambiguity set. Furthermore, a scenario aggregation method is designed to reduce the computational burden. Numerical results for a practical regional power system in China are presented and analyzed to verify the effectiveness of the proposed method.