臺北醫學大學醫學資訊研究所碩士論文摘要 研究所別:臺北醫學大學醫學資訊研究所碩士班 論文名稱:使用改良型基因演算法解決醫師排班問題 指導教授:劉立副教授 研究生:孫雷銘 背景:近年來由於病患安全意識提昇,如何讓值班醫師能有合理適量的工作時間,以減少醫療錯誤的發生,應該會逐漸獲得重視。 目的:以電腦運算自動產生醫師值班表。值班表須符合1.公平性;2. 可因個人需要而調整值班日期;3.無連續值班,或是隔日值班的情形。 材料和方法:以基因演算法為基礎,提出二種改良式交配子(crossover operators),第一種稱為互斥法( repelling);第二種稱為鍵結法( bonding)。另外以四種不同類型的排班需求為測試題,來檢驗傳統的基因演算法與二種改良後的演算法在最佳解、收斂速度、和執行時間上的差異。 實驗結果:每種方法在每個題目上各進行十次,然後求其平均。結果發現互斥法比傳統演算法而言,最佳解進步了7%;收斂速度快了3.5%;執行時間則慢了7.6%,而鍵結法比傳統演算法而言,最佳解進步了30%;收斂速度快了22%;執行時間則慢了2.7%。 結論:鍵結法比傳統的基因演算法不論上收斂速度、最佳解上都有長足的進步,而執行速度只差了一點。主要的原因在於交配子的精細運算,另外也可能得力於二點交配。以後可進行更多點的交配分析及更複雜的排班題目研究。
Background: For the sake of patient safety, the residents’ work hours have got a lot of attentions. It is important to prevent resident physicians working for more than 24 hours to reduce the possibility of accompanying medical errors. Purpose: To generate an on-duty timetable automatically with computers. This timetable should satisfy 1. fairness; 2. individual needs; 3. no consecutive shifts or shifts of every other day. Methods: On the basis of the classical genetic algorithms, two newly devised crossover operators have been proposed. The first one is repelling method; the second one is bonding methods. We tested the two methods with four different kinds of problems. The tests were done on a platform of personal computer and the genetic algorithm Results: Each problem had been tested for each genetic algorithm methods. The results are the average of 10 independent runs. The repelling method is 7% better than classical one in the best individuals; 3.5% quicker in convergent speed; lags 7.6 % in running time. The bonding method is 30% better than classical one in the best individuals; 22% quicker in convergent speed; lags 2.7% in running time. Conclusion: The bonding method overtook both classical and repelling methods in the search of global optima and convergent speed. This may attribute to its effective crossing points selection or its 2-point crossover nature. In the future, we can test multiple-point crossover or uniform crossover and try to solve the more complex scheduling problems.