Human-AI Interactive Framework for Smart Evacuation Safety Analysis in Large Infrastructures

Abbreviation

Pedestrian Evacuation in Large Infrastructures

Large-scale infrastructures such as airport terminals, high-speed railway stations, exhibition centers, stadiums, and subway stations present unique challenges for pedestrian evacuation due to their complex architectural designs, high occupancy levels, and intricate circulation patterns [22]. These facilities are characterized by multiple levels, interconnected spaces, and diverse functional zones that create complex evacuation scenarios significantly different from conventional buildings [23].

Workflow of IEPTool

The function framework and the workflow can be seen in Fig. 2. The Intelligent Evacuation Prediction Tool (IEPTool) is supported by a pre-trained comprehensive AI engine integrated by CNN, LSTM, GAN, and LLM. This engine is trained by a simulation database developed by real architectural floor plans under six types of large infrastructure, including air terminals, exhibition centers, stadiums, high-speed railway stations, metro stations, and passenger stations, respectively. In Chapter 3, the

Evacuation modelling

It should be noted that due to the difficulty of conducting real evacuation experiments in large-scale venues, it is practically challenging to obtain sufficient real evacuation data. Hence, a database of evacuation in various large infrastructures was developed using Pathfinder simulations and real architectural floorplans. Pathfinder is a pedestrian movement analysis tool based on the Social Force Model (SFM), which is widely used and validated in evacuation studies [49,50]. Several key

Performance of the AI model

Fig. 7 shows the predicted performance along different modules of the proposed model. Fig. 7(a) shows that, in the training process, the training strategy of full connection gradually reaches a state of convergence within 100 epochs and achieves a 99% goodness of fitting R². The MAE loss can also indicate a relatively small error in evacuation times. In the testing process, the tolerance range along the testing dataset is between [-2.34%, +2.57%], which means that the predicted evacuation time

Conclusions

This work constructs a simulation database on six types of large space places, including air terminals, large exhibition centers, stadiums, high-speed railway stations, metro stations, and passenger stations, with a total of 66 architectural floor plans. Through data augmentation, the database comprises a total of 264 sets of evacuation time data and corresponding density distribution series, along with 3,076 sequences of exit evacuation flow rates. Additionally, evacuees are categorized into

CRediT authorship contribution statement

Tong Lu: Writing – original draft, Resources, Methodology, Investigation, Formal analysis. Yuxin Zhang: Methodology, Formal analysis, Writing – review & editing, Supervision. Weikang Xie: Resources, Software. Xinyan Huang: Conceptualization, Methodology, Writing – review & editing, Supervision, Funding acquisition.

Data availability

Data will be made available on request.

CRediT authorship contribution statement

Tong Lu: Writing – original draft, Resources, Methodology, Investigation, Formal analysis. Yuxin Zhang: Writing – review & editing, Supervision, Methodology, Formal analysis. Weikang Xie: Writing – review & editing, Methodology, Formal analysis. Xinyan Huang: Writing – review & editing, Supervision, Funding acquisition, Formal analysis, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments