臺灣的住宿式長期照顧(Long-Term Care, LTC)機構火災傷亡數居高不下，但伴隨人口高齡化，LTC服務需求卻仍逐年攀高。就理論來說，雖然住宿式LTC機構存在密集的避難弱勢，這樣的重要設施應在災害的預防跟韌性屬於最高規格。文獻顯示過去雖已增加醫護人員和火災演習次數但仍未有效改善。研究目標認為應改進目前仍使用平面圖與經驗法則的火場搶救佈署。我們從建築資訊模型與本體論出發，發展火場搶救佈署輔助系統。它包含「建築物的物件數值運算」與「搶救指揮官的邏輯客製化」。救火指揮官將設計自己的火場搶救流程，並由系統運算此流程在某建築物的結果。此作法可 1) 減少計算路線所消耗的時間 2) 讓救火指揮官藉由了解機制信任系統的運算結果3) 輕易變換情境設定與指揮官邏輯。 結論認為研究作法不僅提供電腦取代責任制決策的作法也是簡化變更邏輯推論的繁雜作業量。與此同時，現階段臺灣火場救援和建築物結構的弱勢也將在數位化的同時突顯出來。未來希望除了能應用在其他領域外，還能用深度學習取代使用者無法描述邏輯的部分，並分析使用者的信任度變化。;The number of fire casualties in residential long-term care (LTC) facilities in Taiwan remains high, but with the aging population, the demand for LTC services is still rising year by year. In theory, although residential LTC facilities have a dense population of people with disabilities, such critical facilities should be of the highest standard in disaster prevention and resilience. The literature shows that although the number of medical staff and fire drills has been increased in the past, it has not been effectively improved. The goal of the research is to improve the fire rescue deployment that still uses floor plans and rules of thumb. Starting from building information model (BIM) and ontology, we develop a support system for fire rescue deployment. It includes object numerical calculation of buildings and logical customization of fire commander. The fire commander will design his own fire rescue process, and the system will calculate the result of this process in a certain building. This approach can 1) reduce the time consumed in calculating the route 2) let the fire commander trust the calculation result of the system by understanding the mechanism 3) easily change the scenario setting and the fire commander logic. It is concluded that the research method not only provides a method for computers to replace accountability decision-making, but also simplifies the complicated workload of changing logic reasoning. At the same time, the weakness of fire rescue and building structures in Taiwan will also be highlighted as digitization. In the future, we hope that in addition to being applied in other fields, deep learning can be used to replace parts that users cannot describe logic, and to analyze changes in the trust of the user.
