Uncertainty quantifications and operation optimization of buildings as virtual batteries for the grid with high penetrations of renewables

Abstract

In the United States, 40% of the nation’s total energy consumption and 75% of the total electricity consumption are consumed by commercial and residential buildings, and those numbers are projected to be 46% and 80% by year 2035. Primary energy source is limited, and quickly consumed in the modern world. Renewable energy sources such as solar, wind, and biogas have been fully recognized as reliable, environment friendly, and efficient options for low carbon society and sustainable development. A sustainable energy future and net zero energy buildings require high penetration of Renewable Energy Source (RES) such as wind and solar. However, the constrained availability and variability/intermittency of RES present a great challenge to integrate wind and solar energy into the power grid at a large scale. There is a huge potential for building energy cost savings through a proactive integration with the power grid. The key to improve building operation efficiency is to coordinate and optimize the operation of various energy sources with time-sensitive electricity price. In addition, Energy storage systems (e.g., electric battery, thermal storage from chilled water storage tank, etc.) increase the complexity of the whole system integration. Besides the external energy storage equipment such as battery and thermal tank, the building itself also has a hug thermal storage capacity. The thermal mass of buildings has been proved to be able to reduce energy cost and peak demand through precooling strategy. Currently the majority analysis on the building to grid integration system focused on the building demand side management such as how much energy can be saved, and how much operation cost can be reduced. However, the stability of the power grid itself is very important. Currently, the ancillary equipment including flywheels and batteries has high initial cost and operation maintenance fee. Effectively utilizing the existing HVAC equipment and system to provide grid frequency regulation service will help achieve a better building-to-grid integration In order to fully understand the operation of building-to-grid integration system with renewable energy. This dissertation conducted the following research. First, the operation optimization of building-to-grid integration system with renewable energy of solar PV was conducted. Second, the optimal operation schedule of the building with the precooling control strategy using the building thermal mass was investigated. Third, the uncertainty and sensitivity analysis on the operation optimization of the building-to-grid integration system and building with precooling control strategy was performed. . Furthermore, the research on the frequency regulation service from building HVAC system was explored. To solve the proposed optimization problem, there is a need to have accurate forecasting of building energy consumption and solar radiation. Data-driven models such as AutoregRessive model with eXternal inputs (ARX), State Space (SS), Subspace State Space (N4S), and Bayesian Network model (BN) for building energy consumption prediction was evaluated. And recurrent neural network (RNN) based deep learning algorithm for solar radiation prediction was conducted. This dissertation research’s goals are (1) to improve building operation efficiency of building-to-grid integration system by optimizing the operation of various energy sources with time-sensitive electricity price; (2) to improve building operation efficiency with precooling control strategy through building thermal mass; (3) to improve power grid stability by providing frequency regulation service through building HVAC system intelligent operations. Through the dissertation work, the following conclusions were achieved. BN model is more accurate for building energy consumption prediction of all selected data-driven prediction models. The deep learning based RNN model improved the solar radiation prediction performance comparing with the artificial neural network (ANN) model. The accuracy for the ANN and RNN model prediction can be improved with short data sampling frequency and moving window algorithms. Optimization based scheduling for the building-to-grid integration system and the building with a precooling control strategy significantly reduced the building operation cost. Through the uncertainty and sensitivity analysis, it was found that the solar PV area was the most important parameter for the building-to-grid integration system to reduce the operation cost. For the building operation with precooling control strategy, the operation cost is more sensitive to building envelop thermal resistance than the building envelop thermal capacity. Finally, the building intelligent operation also has the contribution to the power stability by providing frequency regulation service.