Efficient building evacuation saves lives during a fire. An inefficiently designed evacuation route may expose occupants to excessive levels of toxic gas, smoke or dangerously high temperatures before they complete their evacuation. This danger is especially high in modern buildings because they usually have complicated layouts and accommodate numerous occupants. Currently, computer simulations have been widely adopted to evaluate the performance of various evacuation systems. Researchers have reported that modelling the detailed movements and interactions among evacuees is critical, as these can greatly affect overall evacuation patterns and times. However, the influence of emotions on local interaction and the entire evacuation process has not been well investigated.

In this study, a novel impatience-based model is proposed for qualitative and quantitative characterization of the impatience level during evacuation. We firstly develop a comprehensive algorithm to calculate the dynamic impatience level considering both the self-growth and the impatience propagation among pedestrians. More specifically, unlike the traditional methods using randomly generated update sequence or fixed sequence, in this model, each pedestrian moves to the target cell in turn for each time step on the basis of the descending order of their impatience level. As a proof of principle, we initially implement it to cellular automata model and perform the parameter sensitivity study of self-growth rate and impatience propagation speed. The result shows that the impatience propagation only promotes the evacuation efficiency when the self-growth rate of impatience is relatively low. When the impatience and patience pedestrians are mixed, the evacuation time can be shortened and then the efficiency can be improved compared with every pedestrian in patience. Note that, any excessive impatience level or insufficient impatience level will lead to an increase in evacuation time. In addition, the investigation towards obstacles shows placing the obstacles strategically can reduce the impatience propagation among pedestrians, and thus improve the evacuation efficiency.

Next, we develop a dynamic impatience-determined route choice model that comprehensively considers the distance and density around exits. We apply this model into three scenarios (i.e., pedestrians are initially in centre, random and non-uniform distributions) and conduct a sensitivity analysis of the self-growth rate and propagation speed of impatience. We find that the centre and random distributions are insensitive to changes in two parameters. Under the non-uniform distribution, the high self-growth rate and low propagation speed of impatience lead to the quickest evacuation process. Moreover, the model enables the simultaneous observation of the pedestrian position and the impatience distribution from the dynamic impatience map. We also conduct a series of experiments to evaluate the model. The experimental and numerical results demonstrate that the impatience level should be limited to an appropriate range because no impatience or excessive impatience leads to increased evacuation times.

Later we present a novel grid-based mesoscopic model for evacuation dynamics. In this model, the evacuation space is discretised into cells larger than those used in microscopic models. This approach directly computes the dynamic changes crowd densities in cells over the course of an evacuation. The density flow is driven by the density-speed correlation. The computation is faster than in traditional cellular automata evacuation models which determine density by computing the movements of each pedestrian. To demonstrate the feasibility of this model, we apply it to a series of practical scenarios and conduct a parameter sensitivity study of the effect of changes in time step δ. Simulation results show that within the valid range of δ, changing δ has only a minor impact on the simulation. The model also makes it possible to directly acquire key information such as bottleneck areas from a time-varied dynamic density map, even when a relatively large time step is adopted. We use commercial software AnyLogic to evaluate the model. The result shows that the mesoscopic model is more efficient than the microscopic model and provides more in-situ details (e.g., pedestrian movement pattern) compared with the macroscopic models.

Lastly, we integrate impatience into the mesoscopic model to investigate the influence of impatience level on the evacuation process. We first test this model in the scenario with a single large exit. Through model parameter analysis by setting time step, initial density, free moving velocities, and impatience level into different values, we find that this impatience-determined mesoscopic model is insensitive towards the change of time step and this model has good stability for predicting the evacuation process even large time step is adopted. Compared with the scenario without consideration of impatience, the dynamic method can short the evacuation time when an appropriate impatience value is adopted since the whole exit can be fully utilized. We further conducted a series of experiments in the classroom with a single exit and two open exits to validate the impatience-determined mesoscopic model. Results of the numerical study are consistent with those of the experiment in terms of evacuation time and the number of pedestrians choosing each exit. It is proved that our impatience-determined mesoscopic model is applicable for predicting the evacuation process. In addition, the time-varied dynamic density map can help the researchers to directly acquire key information such as bottleneck areas and route choice behavior of pedestrians.

Finally, all achievements and conclusions of the work in the thesis are presented and suggestions and outlook for future research are provided accordingly.