Electrical energy flexibilities' prediction and validation of a real non-residential building through methods of machine learning

In times of rising share of renewable energy in the electri-cal distribution grid (EDG), the prediction of electrical en-ergy flexibilities has an important role. Predictions of energy flexibilities of non-residential buildings (NRB) are essential to operate the building in an intelligent and gridfriendly way by storing thermal or electrical energy in the building. The optimal energy storage can be achieved based on the prediction of the NRB's operating conditions in different operation modes. White-box simulation models are commonly used, but the spotlight is shifting to black-box artificial intelligent (AI) models due to their ease of use and the adaptability to seek optimal solutions in rapid manner. In this paper, a black-box and white-box model are created to predict the electrical energy flexibilities of the case study building, the Campus of the Ruhr West University of Applied Sciences (RWU) in Bottrop, Germany. Both models are compared with building's historical measurements to assess their accuracy. The prediction's accuracy of the building's energy flexibilities for both models will be assessed and validated based on a new dataset, which was not considered during the training. A building's potential flexibility scenario is presented for both models. The results show that the AI model offers the same, and sometimes higher accuracy than the white-box model. However, the results reveal that the AI model requires additional training under a flexible operation scheme.