Data-driven coordinated building cluster energy management to enhance energy efficiency, comfort and grid stability

The environmental constraints related to the reduction of emissions and to the contrast to climate change imply an increasing penetration of renewable energy generation. Since renewable energy sources like solar and wind are not programmable and often unpredictable, the grid balancing is becoming more and more challenging. In this scenario, buildings energy flexibility could play a key role through demand side management/load control and demand response actions. The study presented in this thesis work is related to the application of machine learning techniques to the field of building HVAC systems control, in order to enhance their flexibility. The first part of the thesis is related to the study of black box modeling of building thermodynamic behaviour through artificial neural networks. The study is focused on the development of four LSTM models which are able to predict the mean indoor temperature of four different commercial buildings. The novelty lies in the possibility that buildings could not receive their ideal cooling load, but a different one, in order to exploit Demand Response actions. Particular attention is devoted to the process of generation of training dataset, which are created using the EnergyPlus software. The second part of the work is related to the integration of the LSTM models into CityLearn. CityLearn is an OpenAI Gym environment for the implementation of reinforcement learning agents in a multi-agent demand response setting. The aim is to reshape the aggregated curve of electrical demand of the considered buildings: the main novelty of this work is that it would be possible to take advantage of DR actions not only through the control of thermal storages but also through the control of HVAC system Cooling Power, without penalizing thermal comfort.