In order to solve the problem of initial value sensitivity and convergence stability for the wavelet neural network and improve the efficiency and accuracy of the reliability predictive model for metering terminal software, the following steps are performed. The paper improves the whole annealing genetic algorithm (WAGA), and prove that it has extremely strong ability in global convergence and global optimization. Made use of its global optimization property to improve the parameters for wavelet neural network (WNN) and develop model-building method based on whole annealing genetic algorithm-wavelet neural network (WAGA-WNN). Build software reliability predictive model for metering terminal based on the proposed method. The experimental result indicates that this method can solve the problem of initial value sensitivity and convergence stability for wavelet neural network, furthermore, the software reliability predictive model has high efficiency and accuracy.
