Modelling and simulation of hospital evacuation using fuzzy-reinforcement learning based approach [EMBARGOED]

Abstract

Evacuation simulation models are utilized to evaluate and improve evacuation strategies as conducting evacuation drills in health care environments (e.g., hospitals) is extremely difficult. However, available hospital evacuation simulation models usually focus on the movement of the evacuees’ while ignoring the crucial behavioural factors of the evacuees’ which impact the results of the simulation. For instance, the issues of patient prioritization behaviour and exit selection behaviour during evacuation simulation are often overlooked and oversimplified in these models. Moreover, to control the movement of the evacuees, almost all these models utilize rule-based navigation systems which sometimes do not guarantee realistic and optimal movement behaviour. Finding a solution to these limitations is necessary to properly simulate hospital evacuation. Hence, the objective of this research is to develop a hospital evacuation simulation model that addresses the problems related to patient prioritization, exit selection and agent navigation during simulation by modelling feasible and novel solutions to tackle these issues. In this research, we propose to develop a hospital evacuation simulation model to simulate assisted hospital evacuation using the Unity3D where we propose a novel and effective approach to model patient prioritization and exit selection using fuzzy logic controller by considering relevant factors found in the available literature. Moreover, to tackle the issues related to the rule-based navigation system, we propose a reinforcement learning based navigation system where we propose novel reward formulation, observation formulation, action formulation and training procedure. The results and findings exhibited by the proposed model are found to be in line with available literature. For instance, available literature suggests that increased number of patients increases the evacuation time, increased number of staff or exits decreases the evacuation time. Results obtained from the proposed model also demonstrates these findings. Moreover, the proposed navigation system is found to take a “near shortest distance” to reach the target as the mean difference between “shortest vector distance” and “distance covered” are approximately 1.73 m. The proposed simulation model can be used to estimate the Required Safe Egress Time (RSET) which enables the establishment of safety performance levels. The evacuation performance of different scenarios can be compared by using the proposed model. For example, the model can analyse the selection behaviour (e.g., exit selection) of the evacuees in different scenarios. This research can play a vital role in future developments of hospital evacuation simulation models.