Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE credits
Heating, Ventilation, and Air-Conditioning (HVAC) systems constitute a signicant portion of the total energy consumed in households. Changing the operation of the HVAC system can thus have signicant impact on the energy savings that a household can achieve. One way of performing this control is using a model predictive approach, where models of the system are formed, and using these models and their future predictions, an optimal control strategy is found.
This work is then concerned with evaluating di erent models that can predict room temperature changes in a house while the spatial heating system is on and o , as well as models that can predict the energy consumption associated with the heating system usage. The methodology is an improvement over traditional model predictive control, as the models continuously learn over time, improving their results. Data is obtained from sensors placed in 6 NZEBs in Soesterberg, The Netherlands. Black-box models are formed for each house using linear regression, polynomial regression, and a neural network. The models are updated with new information every week so they are able to learn from new data, and are then evaluated based on the magnitude and behavior of their respective errors. Finally the best model for room temperature predictions is found to be a weighted average of the results from the polynomial regression and neural network. A simple linear ordinary least squares model is used for the prediction of energy consumption.
The problem is then formulated as a Markov Decision Process, which allows the system to reduce energy consumption while maintaining user comfort. The genetic algorithm is used to find an optimal control strategy. An optimal control strategy is found with a 24 hour look ahead, while the models take into consideration current weather conditions (also available in the future through weather forecasts) and previous room temperatures. One house was nally taken as an example, where its models were used and an optimal control strategy (a series of set point temperatures) was found for the spatial heating system for every hour over one week in December. The results showed a signicant decrease in energy consumption. The methods used in this work make the loads much more predictable, and allow the exibility o ered by the spatial heating system and the thermal mass of the house to be taken advantage o . The houses at hand are NZEBs and are well designed with small losses, further increasing the potential for energy savings.
2016. , 92 p.