This paper adopts AGP to establish the software reliability
model. The individual of the population according to the case
of the fitness of the generation to adjust the probability of
crossover and mutation by the sigmoid curve, which makes the
crossover and mutation with the direction to convergence and
ensure that the offspring is superior to parents. Therefore, the
development and exploration of the generation is balanceable
and global convergence. By evaluating the data series of the
software testing case in Armored Force Engineering Institute,
the results sufficiently testify that the new AGP algorithm has
better applicability and the validity of fitness and forecasting,
moreover, compared with standard Genetic Programming evolution algorithm, the new AGP algorithm has the better
rapidity of convergence. In a word, the AGP algorithm is
correct, rational and valid, is a more practical algorithm.