Master's Thesis Ken Joo Yeap
Applying data driven cloud computing methods to optimize control of buildings with big thermal inertia.
For a better control of the slow dynamic Thermally Activated Building Systems (TABS), predictive characteristics is a solution to increase the reaction time. This can be done using weather forecast as control input data. Hence, various third party weather forecast services are explored. A Model Predictive Control (MPC) and a predictive approach of the fuzzy control are implemented in this thesis using cloud technologies. The MPC method uses the low ordered Resistance Capacitance (RC) greybox model as the model to ease the modeling effort. Model parameters are generated using a minimization approach of the simulated output of the model and measured data obtained from buildings. With the weather prediction data as the disturbance variable, the model can be used optimize the control taking into consideration future disturbance, allowing the TABS to have more reaction time. The fuzzy rules in the Predictive Fuzzy Control approach are engineered using predesigned experiments based on literature research of advantageous TABS control. Besides, the predictive fuzzy model is constructed to take corrective action during room temperature comfort violation and adaptive measures are eased with logging features. The methods are developed and evaluated using the cloud platform provided by aedifion GmbH on the E.ON ERC main building, which is used as the living lab. Comparing several third party weather forecast data with the on-site weather station, it was found that the DarkSky weather service has a Mean Absolute Error value of less than 2°C for prediction horizon of 0, 1, 6, 24 and 48 hours with hourly coordinate precision weather data. The RC model failed to produce satisfying room temperature results to emulate the E.ON ERC main building due to other disturbances that were not in the scope of this work, especially the internal gain such as occupancy. However, the predictive fuzzy control showed a reduction in energy consumption and an increase in indoor comfort during the first week of application. Finally, an outlook for further research of the MPC and predictive RBC is presented.