地震發生後產生大量傷患，使得醫療資源相對不足，難以在短時間內救治全部的傷患，也使救護車成為稀缺資源，為了減少傷患延誤就醫的情況，有效的規劃救護車派遣，並將傷患送至合適的醫院至關重要。因此本研究參考臺灣消防隊救災流程，以地震發生後多個災點湧出大量傷患為情境，考量災後道路失效的可能，決定救護車派遣與傷患送醫院救治的決策，提出馬爾可夫決策過程（Markov decision process, MDP）模型來求解，該模型使我們能針對當下所發生的情境給予即時且正確的決策，由於問題的複雜度造成MDP 模型的高維度和不可數的狀態空間，使我們難以利用經典的動態規劃求解方法，本研究改用近似動態規劃 （approximate dynamic programming, ADP）求解方法，該求解方法能有效的求出近似最佳決策，我們使用ADP求解技巧來開發近似策略迭代（approximate policy iteration, API）演算法，並使用深度學習（deep learning, DL）進行策略評估，結果與不同啟發式演算法比較皆能獲得有效改善，其中最多能為總完成時間提升117%的效率，期望能給予災害應變人員最佳決策建議。

After the earthquake, a large number of patients are affected, which makes the medical resources relatively insufficient. It is difficult to treat all the patients in a short time. This scarcity of medical resources includes a limited availability of ambulances. In order to improve the survival rate of the patients, it becomes crucial to establish ambulance dispatch and hospital selection policies. Therefore, this research refers to the disaster relief process of the Taiwan Fire Station. We focus on a scenario where numerous patients emerge from multiple disaster sites following the earthquake, and consider the probability of road failures after the disaster. To make immediate and accurate decisions under such circumstances, we construct a Markov decision process (MDP) model. The MDP model aids in determining ambulance dispatch, redeployment, and hospital selection, enabling us to optimize decision-making given the current situation. The complexity of the problem lies in the high dimensionality and uncountable state space of the MDP model, which makes traditional dynamic programming solution methods impractical. Therefore, in this research, we employ an approximate dynamic programming (ADP) solution method. This approach effectively finds approximate optimal decisions. We specifically construct approximate policy iteration (API) algorithm based on ADP solution techniques, and utilize deep learning (DL) for policy evaluation. We hope to provide the best decision-making advice to disaster responders.

摘要 I
Abstract II
目錄 III
圖目錄 V
表目錄 VI
第一章　緒論 1
1.1研究背景與動機 1
1.2研究目的 2
1.3論文架構 3
第二章　文獻回顧 5
2.1動態救護車派遣相關文獻 5
2.2本研究與文獻之差異 7
第三章　數學模型 10
3.1問題定義 10
3.2符號定義 11
3.3數學模型 12
3.3.1系統狀態 13
3.3.2動作（決策） 14
3.3.3狀態轉移機率 14
3.3.4獎勵函數 15
3.3.5目標函數 16
第四章　研究方法 17
4.1近似策略迭代 17
4.2混合代理人和離散事件模擬系統 23
4.2.1模擬系統輸入參數 24
4.2.2離散事件模擬（discrete event simulation） 25
4.2.3代理人模擬（agent simulation） 27
第五章　數值實驗 32
5.1災後情境說明 32
5.2網路參數設定 34
5.3網路搜尋結果 35
第六章　參數分析 40
6.1不同傷患到達時間對最佳解與目標函數值影響 40
6.2不同救護車數量對最佳解與目標函數值影響 42
6.3不同醫院可用病床數量對最佳解與目標函數值影響 44
6.4不同道路損壞程度對最佳解與目標函數值影響 46
第七章　結論與未來研究 49
參考文獻 50

Benson, M., Koenig, K. L., & Schultz, C. H. (1996). Disaster triage: START, then SAVE—a new method of dynamic triage for victims of a catastrophic earthquake. Prehospital and disaster medicine, 11, 117-124.
Jenkins, P. R., Robbins, M. J., and Lunday, B. J., (2021). Approximate dynamic programming for military medical evacuation dispatching policies. INFORMS Journal on Computing, 33, 2-26.
Lam, S. S. W., Ng, C. B. L., Nguyen, F. N. H. L., Ng, Y. Y., and Ong, M. E. H., (2017). Simulation-based decision support framework for dynamic ambulance redeployment in Singapore. International Journal of Medical Informatics, 106, 37-47.
Maxwell, M. S., Restrepo, M., Henderson, S. G., and Topaloglu, H., (2010). Approximate dynamic programming for ambulance redeployment. INFORMS Journal on Computing, 22, 266-281.
Mills, A. F. Argon, N. T. and Ziya, S., (2018). Dynamic Distribution of Patients to Medical Facilities in the Aftermath of a Disaster. Operations Research, 66, 597-892.
Powell, W. B. (2007). Approximate Dynamic Programming: Solving the curses of dimensionality (Vol. 703). John Wiley & Sons.
Jenkins, P. R., Robbins, M. J., & Lunday, B. J. (2021). Approximate dynamic programming for military medical evacuation dispatching policies. INFORMS Journal on Computing, 33, 2-26.
Rettke, A. J., Robbins, M. J., and Lunday, B. J., (2016). Approximate dynamic programming for the dispatch of military medical evacuation assets. European Journal of Operational Research, 254, 824-839.
Robbins, M. J., Jenkins, P. R., Bastian, N. D., and Lunday, B. J., (2020). Approximate dynamic programming for the aeromedical evacuation dispatching problem: Value function approximation utilizing multiple level aggregation. Omega, 91, 102020.
Ruszczyński, A. (2010). Risk-averse dynamic programming for Markov decision processes. Mathematical programming, 125, 235-261.
Shin, K., and Lee, T., (2020). Emergency medical service resource allocation in a mass casualty incident by integrating patient prioritization and hospital selection problems. IISE Transactions, 52, 1141-1155.
Tak, S., Kim, S., and Yeo, H. (2018). Agent-based pedestrian cell transmission model for evacuation. Transportmetrica A: transport science, 14, 484-502.