A New Method of Model Combination Based on the NHPP Software Reliability Models
Haiyan Sun, Lu Zhang, Jing Wu, Ji Tao Wu, Haiyan Yang · 2018
For fitting the failure data with non-steady growth trend better, a new method is proposed to establish combinational models by combining two NHPP models, which have better fitting effects in four candidate models including G-O model, S-Shaped model, M-O model and Duane model for the same data. A real-valued genetic algorithm is used to calculate the maximum likelihood estimates of the parameters in combinational models for solving the problem that the traditional maximum likelihood estimation algorithms cannot give parameter estimates when fitting some data sets. Then, five published failure data sets, most of which have non-steady growth trend, are selected to verify the effectiveness of our models. The results show that the MSE of our models are always lower. It reduces at least 3.12%, and at most 69.31% compared with the single optimal model. These indicate that our method is feasible and effective.