An hierarchical energy system architecture with embedded control will be discussed which pushes prediction models to edge devices to address the challenge of managing demand-side response locally. We employ a two-step approach: At an upper level of hierarchy, we adopt a conventional machine learning pipeline to build load prediction models using automated feature extraction and selection. On a lower level of hierarchy, computed labels are used to train impact models realized by LSTM networks running on edge devices to infer the probability that the power consumption of the player contributes to the upper level prediction failure event. The system is evaluated on individual and aggregated public energy traces of academic buildings.