Ensemble of Markovian stochastic dynamic programming models in different time scales for long term hydropower scheduling

This paper presents a new approach for long term hydropower scheduling. In opposition to the standard Markovian stochastic dynamic programming approach, where monthly inflows are modeled according to probability distribution functions, conditioned to some occurrence of inflow in the previous month, in the proposed approach the monthly inflows are aggregated in different time scales and then submitted to the Markovian model. The discharge decisions are then calculated by a deterministic model that optimizes the problem for one year ahead according to inflows provided by a combination of each Markovian model. Tests were conducted on hypothetical single-reservoirs hydrothermal systems using data from four real Brazilian hydro plants, with distinct hydrological regimes. The performance of the proposed method was evaluated through simulation, using the historical inflow data, in comparison with the standard Markovian model. The results have shown that the proposed approach has provided spillage reduction and increase on hydro productivity as well as power generation, which incurred in up to 2.1% reduction in operational costs.