Wind power prediction interval forecasting using DeepAR and Deep State models.

The greatest issue in combining wind power to electricity grid is its intermittent of wind resources. Accurate wind power generation forecasts are an effective technique for dealing with such an issue. Wind power forecasting (WPF) generates the conditional expectation of wind power generation (WPG) at certain times. Any prediction carries with it an element of risk. Various probabilistic forecasting approaches in wind energy forecasting are analyzed. Probabilistic predictions opposed to the point forecasts that offer huge quantitative information on the uncertainty related with wind power generation. Probabilistic projections are the best inputs for making decisions in an uncertain world. In this work, Prediction interval forecasting is performed using DeepAR and DeepState models. This work uses wind data from wind turbine (10-minute) time interval (TURKEY). The wind dataset has five attributes and data is captured every ten minutes. The investigation's ultimate goal is to predict the wind speed prediction interval. The experiment is carried out in GluonTS framework to compute the prediction Interval. Multilayer perceptron and DeepState Space models are used, and its performance is measured using PCIP and PINAW metrics. [ABSTRACT FROM AUTHOR]