Electricity Consumption Forecast of Clusters of Buildings Based on Recurrent Neural Networks

During the last decade, the relevance of improving the energy efficiency and thus reduce the energy consumption of buildings has gained momentum for many reasons. In addition to economic and sustainability reasons, an important factor is to ensure and maintain comfortable and healthy conditions inside buildings. Indeed, the study of the behavior of users inside buildings is essential to ensure comfortable and healthy conditions in living environments and cannot be avoided when defining energy saving measures. To achieve this goal, this paper presents a simulation framework for the prediction of the electricity consumption of cluster of buildings based on users' behaviors. The framework is based on model-based approaches simulating the energy consumption of buildings and statistical models representing the behavior of users. Simulated energy consumption profiles are then used to train recurrent neural networks that, based on real energy consumption data, can be used to tune the statistical and deterministic parameters of the simulation models. University campuses comprising different type of buildings are used as reference use case in this paper, as representative example of energy cluster of buildings.