在資訊科技產業中,精準的評估每個軟體開發專案的工作量對於軟體公司在軟體開發成本與開發時程管理是非常重要的。從一個專案開始,多數的開發團隊都會覺得時間不夠用或是專案評估錯誤導致軟體專案失敗。然而軟體專案的成本幾乎都是人力成本,人力成本又與開發時程成正比,所以精準的工作量評估就更顯得重要了。因此 ,本研究將以華德法分群演算法(Ward method clustering algorithm) [Ward 1963]對專案進行分群,再將分群後的專案,透過差分演化演算法 (Differential evolution algorithm)以平均誤差率(MMRE)做為適應值不斷的往最佳參數趨近,最後再利用最佳化後的參數引用至欲做預測的專案求出軟體工作量。本研究利用COCOMO 中的63筆歷史專案來進行測試,實驗結果確實表現出利用多因子做為專案分群依據能比COCOMO最初的三個模式更有效的預估軟體工作量。
In Information Technology Industry, how to accurately estimate one project’s spending in the cost and works always plays a very important role in software companies. Therefore, this research via Ward method clustering algorithm and applies differential evolution, a new algorithm, to estimate the optimal volume of works in software projects to acquire biggest benefit in the cost. In algorithm, generations of chromosomes firstly use Pred and MMRE to figure out their fitness then keep renewing their values by Mutation, Crossover, and Selection to substitute the older and worse vector solutions. In the process of repeating these technique, finally, vector solutions will be refine to their optimal volume which can be used in estimating software projects.