Comparing R-Vine Copulas and Quantile Regression Forests for Reliability Forecasting of Renewable Energies

This work presents the comparison of R-Vine Copulas (RVCs) and Quantile Regression Forests (QRF) used to forecast the reliable power output of a pool of wind power plants (WPPs) and solar power plants (SPPs). Both methods have been used to forecast the probability of deviations from the deterministic power forecast of WPPs and SPPs but have never been compared using the same basic data set for the parameterization and validation. RVCs are a mathematical model for the description of stochastic dependencies, while QRF is a data-driven machine learning model, which describes the relation between one or more predictor variables and a response variable by using a combination of decision-trees. Both methods will be explained briefly and applied to predict the power of different exemplary pools of WPPs and SPPs with the same basic weather data set. Our paper contributes by giving a quantitative comparison in reliability and performance of the generated forecasts as well as formulating recommendations for the use cases of both tools. The results show that QRF forecast with a higher reliability but are more limited in their use case then RVCs.

Subjects

forecasting

Quantile Regression Forests

R-Vine Copula

Renewable Energies

stochastic dependencies

MLE@TUHH

Options

Comparing R-Vine Copulas and Quantile Regression Forests for Reliability Forecasting of Renewable Energies