Parameter optimization of software reliability models using improved differential evolution algorithm

Abstract

Differential evolution (DE) is known as a strong and simple optimization method able to work with non-differential, nonlinear, and multimodal functions. This paper proposes a modified differential evolution (MDE) algorithm for solving a high dimensional nonlinear optimization problem. The issue is finding maximum likelihood estimation (MLE) for the parameters of a non-homogeneous Poisson process (NHPP) software reliability model. We make two modifications to DE: a mutation scheme based on a new affine combination of three points for increasing the exploration power of the algorithm, and another is a uniform scaling crossover scheme to increase the exploitation ability of the algorithm. The performance of the proposed scheme is empirically validated using five software reliability models on three software failure datasets. Analysis of research findings indicates that the proposed scheme enhances the convergence speed of the DE algorithm, and preserves the quality of the solution. A comparison with two other peer algorithms is also shown the superiority of the proposed algorithm.