Deep learning-based multi-output quantile forecasting of PV generation

This paper develops probabilistic PV forecasters by taking advantage of recent breakthroughs in deep learning. A tailored forecasting tool, named encoder-decoder, is implemented to compute intraday multi-output PV quantiles forecasts in order to efficiently capture the time correlation. Quantile regression, a non-parametric approach, is selected as it assumes no prior knowledge of the probabilistic forecasting distribution. The case study is composed of PV production monitored on site at the university of Liège (ULiège), Belgium. The weather forecasts from the regional climate model provided by the Laboratory of Climatology of the university of Liège are used as inputs of the deep learning models. The quality of the forecasts are quantitatively assessed by the continuous ranked probability and winkler scores. The results indicate this architecture improves the forecast quality and is computationally efficient to be incorporated in an intraday decision making tool for robust optimization.