J-M模型概率逼近参数估计方法

Probability Approximation Method to J-M Model Parameter Estimation

指出了J-M模型传统算法在进行参数估计时存在的不足,对J-M模型进行了深入分析,发现了模型中软件失效时间的概率分布特性,建立了n重贝努利概型,首次提出了精确的J-M模型参数估计目标,最先设计并使用了概率逼近的参数估计算法,并在标准的软件可靠性失效数据一海军战术数据系统(NTDS)上对该算法进行了验证。与传统的基于极大似然估计法和最小二乘法的参数估计方法相比,新算法充分考虑了软件失效时间的随机性,更符合概率统计规律,不受极限条件限制,简捷高效,收敛迅速,估计结果精度有了质的改变,更加符合实际,具有很好的应用价值。

The deficiency of the traditional parameter estimation algorithm of J-M model was pointed out, the probability distribution features of software failure time was found via a deeper analysis of J-M model, the Bernoulli model was built, the precise target of J-M model parameters estimation was proposed, probability approximation parameters estimation algorithm was designed and used first time, and the algorithm was validated via the standard software reliability failure data of Navy Tactic Data System （NTDS）. Compared to traditional parameter estimation algorithm based on maximum likelihood estimation and least square method, the new algorithm takes full account of the randomness of the software failure time and it is more in line with the laws of probability and statistics, not subject to extreme conditions, simple and ejficient, furthermore, it has more rapid convergence. The accuracy of estimation results has a qualitative change and the results are more in line with reality. The parameter estimation method has very good application value.