Speaker
Description
As the complexity of urban disaster management increases, the construction of urban disaster-related ontologies becomes crucial. However, existing urban disaster-related ontologies do not sufficiently consider human factors, particularly in the knowledge modeling of dynamic interactions among humans, disasters, and the environment in disaster response, making it difficult to effectively support disaster prediction and emergency decision-making. Therefore, this study develops a human-centered ontology framework that integrates deep reinforcement learning (DRL) and multi-agent reinforcement learning (MARL) to model the evolving interactions among humans, disasters, and urban environments.
First, this study explores the dynamic updating method of the ontology when introducing real-time human interaction data, analyzing the applications, challenges, and limitations of deep reinforcement learning and multi-agent reinforcement learning in ontology updates. Then, the proposed ontology adopts a multi-layered structure, designed through expert validation and OWL 2 DL reasoning, ensuring logical consistency and adaptability. It incorporates task-specific roles (e.g., first responders, emergency managers, vulnerable populations), human physiological and psychological states (e.g., stress-induced decision biases, fatigue accumulation), and human-centric IoT data (e.g., biometric sensors, wearable devices, mobility traces), among others. By capturing these factors, the ontology enhances disaster situation awareness, enabling more accurate risk assessment and adaptive decision-making.
Finally, this study evaluates the applicability of the ontology using a typical open-source disaster database. Preliminary validation in disaster simulations demonstrates that the ontology can dynamically adjust evacuation strategies based on smart city digital twins and multi-hazard scenarios.
This study constructs a human-centered urban disaster ontology, optimizing disaster information representation, knowledge reasoning, and emergency management, thereby enhancing the intelligence level of disaster management. The ontology can be integrated into a smart city disaster management system, supporting cross-hazard risk assessment and resource scheduling optimization, and promoting the transformation of disaster management from static knowledge bases to dynamic cognitive systems.
| Keywords | Human-centered ontology; Reinforcement learning; Dynamic adaptation; Urban disaster; Emergency decision-making |
|---|---|
| Best Congress Paper Award | Yes |
