PERFORMANCE AND APPLICATIONS OF RESIDENTIAL BUILDING ENERGY GREY-BOX MODELS
Siemann, Michael James
MetadataПоказать полную информацию
The electricity market is in need of a method to accurately predict how much peak load is removable by directly controlling residential thermostats. Utilities have been experimenting with residential demand response programs for the last decade, but inconsistent forecasting is preventing them from becoming a dependent electricity grid management tool. This dissertation documents the use of building energy models to forecast both general residential energy consumption and removable air conditioning loads. In the models, complex buildings are represented as simple grey-box systems where the sensible energy of the entire indoor environment is balanced with the flow of energy through the envelope. When internet-connected thermostat and local weather data are inputs, twelve coefficients representing building parameters are used to non-dimensionalize the heat transfer equations governing this system. The model's performance was tested using 559 thermostats from 83 zip codes nationwide during both heating and cooling seasons. For this set, the average RMS error between the modeled and measured indoor air temperature was 0.44°C and the average daily ON time prediction was 1.9% higher than the data. When combined with smart power meter data from 250 homes in Houston, TX in the summer of 2012 these models outperformed the best traditional methods by 3.4 and 28.2% predicting daily and hourly energy consumption with RMS errors of 86 and 163 MWh. The second model that was developed used only smart meter and local weather data to predict loads. It operated by correlating an effective heat transfer metric to past energy data, and even further improvement forecasting loads were observed. During a demand response trial with Earth Networks and CenterPoint Energy in the summer of 2012, 206 internet-connected thermostats were controlled to reduce peak loads by an average of 1.13 kW. The thermostat building energy models averaged forecasting the load in the 2 hours before, during, and after these demand response tests to within 5.9%. These building energy models were also applied to generate thermostat setpoint schedules that improved the energy efficiency of homes, disaggregate loads for home efficiency scorecards and remote energy audits, and as simulation tools to test schedule changes and hardware upgrades.