In this talk we discuss the application of reinforcement learning for more sustainable energy systems. In the first part of the talk, we look at increasing the efficiency at the supply side; specifically, from wind turbines. We show results, on real-world data and industry benchmarks, of modest but significant gains over standard controllers. We also note that increasing the capacity at the supply side only makes sense if matched closely to demand, yet fluctuations are inherent to both supply (especially from renewable sources such as wind) and demand (i.e., consumers). Therefore, we study means to model the dynamic preference of consumers from observed energy use, via inverse reinforcement learning, in a non-intrusive way. The goal is to understand the dynamics of behavior, such that it may be included as an additional objective in matching demand to supply; and overall, moving towards a more sustainable energy system in the long term. This talk is based heavily on the work of Alban Puech and Simo Alami stemming from their collaboration with DEIF Wind Power Technology and EDF France (via SystemX), respectively.