以擴散式粒子群最佳化為基礎之智慧型多目標 護士排程系統

Using Scatter PSO for Intelligent Multi-Objective Nurse Scheduling

醫院為了兼顧管理營運、經營利益、政府法規、和護理人員排班公平性等因素，在傳統人工作業模式下，需要花費很多時間才能產生合理的排班表，但又無法客觀評估各種目標的達成效益。本論文提出 多目標擴散式粒子群最佳化演算法（Scatter MOPSO）結合禁制搜尋法來解決護士排班問題，醫院經營者 可擬訂多個營運目標（如降低成本及提高員工滿意度等），並述明排班限制（如符合營運實務、職場安全限制、及政府法規等），利用本論文提出的數學規劃模式即可自動求得既符合排班限制又可同時優化多個 目標的一組解集合，形成 Pareto front，進而提供醫院經營者可量化的決策參考依據。實驗結果顯示，在 護士排班測試問題上，我們所提出的方法 Scatter MOPSO 比文獻上的 MOPSO 得到更佳的排班品質。最後，我們分析 Scatter　MOPSO 的收斂行為，在不同條件的護士排班測試問題中，皆顯示能快速而有效的 趨近最佳解鋒面，同時也能逐步產生更多彼此分佈均勻的不被支配解。

It is time-consuming to generate a nurse scheduling using traditional human-involved manner in order to account for administrative operations, business benefits, governmental regulations, and fairness perceived by nurses. Moreover, the objectives cannot be measured quantitatively even when the nurse scheduling is generated after a lengthy manual process. This paper presents an Multi-Objective Scatter PSO combined with Tabu Search to tackle the real-world nurse scheduling problem. By the proposed mathematical formulation, the hospital administrator can set up multiple objectives (such as cost reduction and nurse-satisfaction raising) and stipulate a set of scheduling constraints (such as operational practice and governmental regulations), and our system can automatically generate a set of solutions which nearly optimize the given objectives and meet the specified constraints. The experimental results manifest that our method (Scatter MOPSO) performs better than MOPSO on benchmark nurse scheduling problems.