Machine learning for modeling energy systems complexity

In the context of rising concerns on climate change, efforts are made to improve our energy systems with the target of a carbon-neutral society by 2050. Energy systems are complex systems yet most modeling approaches do not consider their core aspects described by complexity science so far. These aspects can be grouped into Adaptation, Nonlinearity, Self-organization, and Network, which are interrelated and without which one cannot explain energy systems complex dynamics. The present study suggests the use of machine learning models to capture energy systems complexity. Some studies have already implemented machine learning for cluster analysis or forecasting purposes, but the potential of such techniques is still far from being fully exploited. We remind some understanding of complex systems then explain how machine learning models can complement or surpass current models as well as their limitations. Concrete examples are given for real applications.