A hybrid evolutionary probabilistic forecasting model applied for rainfall and wind power forecast

Several works in the literature so far have been focused on deterministic point forecasts, which, usually, indicates the conditional mean of future observations. An increasing need for generating the entire conditional distribution of future observations has been required for the new generation of soft sensors. This study aims the probabilistic forecasts, reporting the use of a hybrid fuzzy forecasting model applied in two different forecasting problems. Our adapted model is applied to predict the rain of the city of Vitoria, in the state of Espírito Santo, Brazil. Real data from a wind farm, provided by the Irish EirGrid institute, was used for analyzing the proposal over a real time series with high fluctuations. Due to the stochasticity of the the hybrid model, which is calibrated through the use of an evolutionary metaheuristic, we adapted it in order to generate future using quantile regression. Computational experiments indicated the ability of the model in finding useful probabilistic quantiles, which were flexible enough in order to limit the lower and upper bounds of the historical datasets. While the probabilistic quantiles suggested the probability of rain and its magnitude, they were also able to predict expected ranges of the amount of energy generated from the wind farm.