No abstract

No abstract

Energy systems are undergoing a profound transition worldwide, substituting nuclear and thermal power with intermittent renewable energy sources (RES), creating discrepancies between the production and consumption of electricity and increasing their dependence on greenhouse gas (GHG) intensive imports from neighboring energy systems. In this study, we analyze the concurrent electrification of the mobility sector and investigate the impact of electric vehicles (EVs) on energy systems with a large share of renewable energy sources. In particular, we build an optimization framework to assess how Evs could compete and interplay with other energy storage technologies to minimize GHG-intensive electricity imports, leveraging the installed Swiss reservoir and pumped hydropower plants (PHS) as examples. Controlling bidirectional EVs or reservoirs shows potential to decrease imported emissions by 33–40%, and 60% can be reached if they are controlled simultaneously and with the support of PHS facilities when solar PV panels produce a large share of electricity. However, even if vehicle-to-grid (V2G) can support the energy transition, we find that its benefits will reach their full potential well before EVs penetrate the mobility sector to a large extent and that EVs only contribute marginally to long-term energy storage. Hence, even with a widespread adoption of EVs, we cannot expect V2G to single-handedly solve the growing mismatch problem between the production and consumption of electricity.

Replacing poorly performing existing controllers with smarter solutions will decrease the energy intensity of the building sector. Recently, controllers based on Deep Reinforcement Learning (DRL) have been shown to be more effective than conventional baselines. However, since the optimal solution is usually unknown, it is still unclear if DRL agents are attaining near-optimal performance in general or if there is still a large gap to bridge.In this paper, we investigate the performance of DRL agents compared to the theoretically optimal solution. To that end, we leverage Physically Consistent Neural Networks (PCNNs) as simulation environments, for which optimal control inputs are easy to compute. Furthermore, PCNNs solely rely on data to be trained, avoiding the difficult physics-based modeling phase, while retaining physical consistency. Our results hint that DRL agents not only clearly outperform conventional rule-based controllers, they furthermore attain near-optimal performance.