A non parametric stochastic model for river inflows based on kernel density estimation

One of the most important random variables to be considered in hydrothermal systems operation and planning models is river inflows, especially in countries with a very high penetration of hydro power plants. Usually inflows are represented by periodic autoregressive parametric models, assuming that they follow a log-normal distribution. This paper proposes the application of a non-parametric estimation method, called kernel density estimation, for the construction of an autoregressive model for river inflows that does not assume any specific distribution. The synthetic series generated by the proposed model tend to better reproduce the probability density of the historical time series. The model was applied to some hydrolological series of the Brazilian system, and its efficiency was demonstrated as well as some advantages over the conventional approach.