Scenario Development Support for Disaster Response Headquarters Tabletop Exercises Using Disaster Response Verification Reports and Large Language Models

Abstract

In tabletop exercises conducted by disaster response headquarters to improve disaster response capabilities, realistic and diverse scenarios are essential for enhancing decision-making and coordination with relevant agencies. However, conventional scenario development relies heavily on the experience and knowledge of the authors, presenting challenges in terms of the considerable time and effort required. In this study, we developed a scenario generation support system utilising a large language model trained on past disaster response verification reports. By converting the report into a question-and-answer format dataset and performing additional training, we made it possible to present examples of incidents, responses, and challenges tailored to specific training conditions as scenario material. The results of the evaluation experiments confirmed the potential of this approach to assist controllers in scenario creation and to diversify scenario content.