Deterministic versus stochastic models for long term hydrothermal scheduling

Abstract

This paper compares the performance of deterministic and stochastic models for long term hydrothermal scheduling. Two deterministic models are taken into consideration: a deterministic dynamic programming model based on average inflows; and a model based on tracking a seasonal storage curve defined by average values provided by a deterministic nonlinear programming model on inflow historical records. As stochastic models, two dynamic programming approaches are considered: the first one represents the inflow by independent probability distribution functions; and the second one adopts a dependence of lag-one through periodical autoregressive models. In order to concentrate the analysis on the stochastic aspect of the problem, the case studies performed have considered single reservoir systems. The performance comparison was based on statistics of mean and standard deviation of generation and cost obtained by simulating the operation on the inflow historical records. Three hydropower plants located in different river basins in Brazil were selected for the case studies.