Abstract

Forecasting simulations of building environment have attracted growing interests since more and more applications have been explored. Occupant’s thermal comfort, safety and energy efficiency are reported to directly benefit from accurate predicted building physical conditions. Among all available research regarding forecasting indoor environment, there are substantially fewer studies relating to occupant safety and emergency forecasting and response than that of comfort and energy savings. This may due to the nature that the forecasting simulations associated with life safety concerns demand higher accuracy. Although the tasks of forecasting potential threats in the indoor environment are especially challenging, the benefits can be significant. For example, toxic contaminants such as carbon monoxide from fire smoke can be monitored and removed before the concentration reaches a harmful level. The sudden release of hazardous gases or the smoke generated from an accidental fire can also be detected and analyzed. Then, based on the results of forecasting simulations, the building control system can provide an efficient evacuation plan for all occupants in the building. However, by using traditional simulation tools that utilize one set of initial inputs to forecast future physical states, the predicted physical conditions may depart from reality as the simulation progresses over time.

In this thesis, forecasting simulations of building safety management are improved by applying the theory of data assimilation where the simulation results are aided by the sensor measurements. Instead of studying methods that require high computational resources, this research focuses on affordable approaches, ensemble-based algorithms, to forecast indoor environment to solve various safety problems including forecasting indoor contaminant and smoke transport. The resulting models are able to provide predictions with noticeable accuracy by only using affordable computer resources such as a regular PC. Finally, a scaled compartment fire experiment is conducted to verify the real-time predictability of the model. The results indicate that the proposed method is able to forecast real-time fire smoke transport with significant lead time. Overall, the method of Ensemble Kalman Filter (EnKF) is efficient to apply to forecasting indoor contaminant and smoke transport problems. In the end of this thesis, suggestions are summarized to help those who would like to apply EnKF to solve other building simulation problems.