Deep Quantile Regression Based Wind Generation and Demand Forecasts

The widespread attention in the growth of clean energy for electricity production necessitates an accurate and reliable generation and demand forecasts. However, the decision-making process in electric power industry involves more uncertainty due to the transition towards distributed energy systems, which are not addressed in the conventional point forecasts. This paper proposes a probabilistic method termed as Deep Quantile Regression (DQR) for the construction of prediction intervals (PIs) that can potentially quantify uncertainty in the point forecasts of wind power generation and demand. The effectiveness of DQR is examined using the low and high seasonal wind and demand datasets. PIs with various confidence levels of 99%, 95% and 90% are estimated by constructing the appropriate quantiles using the proposed DQR method. The quantitative comparison of the quality in all the estimated PIs using the proposed method proves to outperform the other state-of-the-art methods.