AI for Energy: Data-efficient and explainable algorithms to exploit energy flexibility

Chris gave an invited talk on the Sustainable Electric Power Generation under EU Conditions workshop (https://kee.fei.tuke.sk/?page_id=2297), highlighting a number of AI techniques that will be used, improved and validated in BlueBird to make buildings truly flexible assets.The talk discussed:- Physics-informed reinforcement learning as a way to learn smart control policies from limited data sets by infusing a priori physics knowledge into neural networks.- Distillation of interpretable decision trees from black-box policies as a solution to improve the transparency of AI based control systems.- The HomeLabGym python library which offers a standardized interface to easily validate new control policies in HomeLab, a real house used as living lab environment.