Xiwang Li

Building Energy Consumption On-Line Forecasting Using System Identification and Data Fusion

Model based control has been proven to have significant building energy saving potentials through operation optimization. Accurate and computationally efficient, and cost-effective building energy model are essential for model based control. Existing studies in this area have mostly been focusing on reducing computation burden using simplified physics based modeling approach. However, creating and identification the simplified physics based model is often challenging and requires significant engineering efforts. Therefore, this study proposes a novel methodology to develop building energy estimation models for on-line building control and optimization using an integrated system identification and data fusion approach. System identification model has been developed based on frequency domain spectral density analysis. Eigensystem realization algorithm is used to generate the state space model from the Markov parameters. Kalman filter based data fusion technique has also been implemented to improve the accuracy and robustness of the model by incorporating with real measurements. A systematic analysis of system structure, system excitation selection as well as data fusion implementation is also demonstrated. The developed strategies are evaluated using a simulated testing building (simulated in EnergyPlus environment). The overall building energy estimation accuracy from this proposed model can reach to above 95% within 2 minutes calculation time, when compared against detailed physics based simulation results from the EnergyPlus model.Copyright

Commercial building cooling energy forecasting using proactive system identification: A whole building experiment study

Model-based predictive control has been proven to be a promising solution for improving building energy efficiency and building-grid resilience. High fidelity energy forecasting models are essential to the performance of model predictive controls. The existing energy forecasting modeling principles: physics based (white box), data-driven (black box), and hybrid (gray box) modeling principles all have their own limitations in applying into the real field, such as extensive engineering effort, computation power, and long training periods. Previous studies by the authors presented a novel methodology for energy forecasting model development using system identification approaches based on system characteristics. In this study, whole building experiments are systematically designed and conducted to verify and validate this novel method in a real commercial building. The experimental results demonstrate that the proposed methodology is able to achieve 90% forecasting accuracy within a 1-minute calculation time for chiller energy and total cooling energy forecasting in a 1-day forecasting period under the experimental conditions. A Monte Carlo study also shows that the model is more sensitive to outdoor air temperature and direct solar radiation, but less sensitive to ventilation rate.