本論文研究軟體在指定的時間間隔內具有多重失效性質的軟體可靠度成長模型估計。 本論文採用累積過程來預估, 直到時間 t 為止, 每一次軟體失效所產生的錯誤或故障數目。 在本文內所提出的累積過程模型, 可以用來處理在相同 CPU 計算單位或單位時間內, 所有造成軟體故障的集群的狀況。 假設軟體失效到達過程為一種非齊性卜瓦松過程, 而每一次失效發生時, 錯誤或故障數目則服從特定的複合分佈。 其中, 幾何分佈和對數分佈分別具有故障無記憶性和傳染性的特性。 我們綜合非齊性卜瓦松過程和複合分佈, 來估計樣本在累積過程下的平均失效數。 根據文中所提出的模型, 我們也藉由一些訊息準則進行模型的驗證。 例如,AE、 RE、 K-S 等。 針對軟體可靠度所建立的措施, 可以根據以上的公式來計算。 當軟體具有故障集群的存在時, 運用累積過程的非齊性卜瓦松模型進行的分析, 明顯優於單純考慮非齊性卜瓦松的模型。 This thesis studies the estimation of software reliability growth models when there are multiple failures that occur simultaneously in groups within the specified time interval. A cumulative process is proposed for estimating the number of failures up to testing time t. The proposed cumulative process model can be used to deal with the situation that software failures cluster within the same CPU second or unit time. The failures arrival process is assumed to be non-homogeneous Poisson processes, and the cram sizes are governed by specific compounding distributions.The geometric and logarithmic distributions are respectively taken when failures are memoryless and contagious. The mean value of the number of failures of the cumulative process is estimated by the sample path functions of arrival non-homogeneous Poisson process and compounding distribution. Performances of the proposed models are verified by some information criteria. The software reliability measures are then computed according to their formulas. The cumulative process models indicate that they are superior to ordinary NHPP models where clustering of failures exists.
