Outlier Detection in Energy Climate Data

Abstract

The energy transition, moving away from fossil fuels to renewable energy sources, introduces an increasing variability in energy generation. In order to prepare for the future, signi?cant improvements need to be made to the energy grid. We investigate the use of outlier detection algorithms to improve the assessment of future energy systems. Outliers represent critical conditions that should be taken into consideration by policy makers when designing the future energy grid. We combine the MDI algorithm and the SLOM algorithm with novel post-processing to detect temporal and spatial-temporal outliers. These algorithms are applied to energy generation data derived from ERA5 historical climate reanalysis data using energy conversion models. Using the MDI algorithm we found temporal outliers that are potential risks for the energy grid. We found that the application of SLOM, a spatial outlier detec- tion algorithm, and the post-processing, provided no new insights. Historical trends that could be attributed to climate change were investigated but not found. For the historical period we found that outlier intensity might be in u- enced by multidecadel variability. We conclude that our method shows that outlier detection might help the assessment of the future energy grid by high- lighting the most extreme situations. Researchers and policy makers could use information on the discovered outliers to improve the future development of the electricity network.