Physics-enhanced machine learning for energy systems

Abstract:

Building operations account for a huge amount of energy usage and the HVAC (Heating, Ventilation and Air Conditioning) systems are the largest consumer of energy in this sector. A way to reduce this demand is to implement more effective control algorithms and a popular way to do this is by using model based control strategies. However, this demands a precise model. Modelling heat in buildings is a difficult task since a lot of disturbances occur in the process. Computers running, lots of people in a room and an open window are all examples of disturbances that exists and have to been taken into account.

This thesis aims to use recent data-driven methods to find suitable procedures to model heat in buildings. This was done in two steps. First, a gray box model was created and its parameters fitted using different data-driven methods. Then, more complex learning-based models were tried out and added to the gray box part to catch some of the disturbances. Feed-forward neural networks, LSTM networks and box jenkins models were the methods used for this disturbance modelling part.

The results showed that a gray box model can capture most of the dynamics of the heat dynamics in the building, but that the obtained parameters and the performance depended a lot of the method used for parameter estimation. Adding a more complex disturbance part to the gray box model improved the results by far and were able to catch some of the disturbances.