EVOX-CPS : A Methodology For Data-Driven Optimization Of Building Operation

Existing building stock’s energy efficiency must improve due to its significant proportion of the global energy consumption mix. Predictive building control promises to increase the efficiency of buildings during their operational phase and thus lead to a reduction of the lion’s share of buildings’ lifetime energy consumption. Predictive control complements other means to increase performance, such as refurbishments as well as modernization of systems. This thesis contributes EVOX-CPS, a holistic methodology to develop data-driven predictive control for (existing) buildings and deploy the control in day-to-day use. EVOX-CPS evolves buildings into Cyber-Physical Systems and addresses the development of data-driven predictive control using computational methods. The thesis’ focus rests on accounting for the situation of existing buildings - which vary greatly regarding their physical characteristics, usage patterns, system installation, and instrumentation levels. The methodology addresses the aspect of building stock variety with its capability to flexibly adapt to different buildings’ characteristics, e.g., by supporting the integration of varying levels of pre-existing building instrumentation. Furthermore, EVOX-CPS supports using different data mining, regression, or control techniques (i) to strengthen the support for a variety of buildings, and (ii) to cater to researchers’ and practitioners’ differing skills, experiences, or preferences concerning different data analysis techniques. Through its flexibility, the methodology addresses a vast potential installation base and lowers the barriers for adoption in day-to-day use, e.g., by being able to leverage prior investments in building instrumentation and supporting different data-analysis techniques. At the same time, EVOX-CPS provides researchers and practitioners with comprehensive guidance relevant to their daily work. Besides, EVOX-CPS supports addressing a building’s known limitations in the daily operation, e.g., uncomfortable indoor conditions. The experimentation in two real buildings validates the effectiveness of EVOX-CPS’ data-driven control with high reliability due to prolonged experimentation periods combined with applying energy normalization and inferential statistics. The experiments during routine heating system operation establish high confidence in the recorded effect sizes: the improvements in operational efficiency are profound and statistically significant. More specifically, the experiments of controlling the grass heating system of the soccer stadium Commerzbank Arena, Frankfurt, Germany, in two winters saved up to 66% (2014/2015) and 85% (2015/2016) of energy consumption. Extrapolation to an average heating season leads to expected savings of 775 MWh (148 t of CO2 emissions) and 1 GWh (197 t CO2), respectively. The experiments also show that EVOX-CPS allowed alleviating the known operational limitation of heating supply shortages which required nightly preheating in the stadium’s standard operating procedures. In another set of experiments, we applied the methodology to control the heating system of the Sierra Elvira School in Granada, Spain. The experimentation occurred during the regular class hours of 43 school days in winter 2015/2016. A first experiment demonstrated the possibility to lower consumption by one-third while maintaining indoor comfort. Another experiment raised average indoor temperatures by 2K with 5% additional energy consumption. Again, that illustrates EVOX-CPS’ capability to address a building’s known operational issues