Building occupancy simulation and estimation simulates the dynamics of occupants and estimates the real time spatial distribution of occupants in a building. It can benefit various applications like conserving energy, smart assist, building construction, crowd management, and emergency evacuation. Building occupancy simulation and estimation needs a simulation model and a data assimilation algorithm that assimilates real-time sensor data into the simulation model. Existing build occupancy simulation models include agent-based models and graph-based models. The agent-based models suffer high computation cost for simulating a large number occupants, and graph-based models overlook the heterogeneity and detailed behaviors of individuals. Recognizing the limitations of the existing models, in this dissertation, we combine the benefits of agent and graph based modeling and develop a new graph based agent oriented model which can efficiently simulate a large number of occupants in various building structures. To support real-time occupancy dynamics estimation, we developed a data assimilation framework based on Sequential Monte Carol Methods, and apply it to the graph-based agent oriented model to assimilate real time sensor data. Experimental results show the effectiveness of the developed model and the data assimilation framework. The major contributions of this dissertation work include 1) it provides an efficient model for building occupancy simulation which can accommodate thousands of occupants; 2) it provides an effective data assimilation framework for real-time estimation of building occupancy.