Quantum direct current electricity meter is one of the important instruments in smart grid, the reliability growth model is of great significance to improve its reliability. In the past, when several types of commonly-used neural networks were used for modeling, there were problems like low parameter training efficiency and low generalization ability caused by unsatisfactory parameters, which reduced the prediction accuracy of the models to a certain extent. In this paper, we will replace the training process of the neural network with a parameter optimization process, and use the improved whole annealing genetic algorithm (WAGA) to optimize the parameters of the back propagation neural network. This improves the modeling efficiency by 18 times and significantly improves global optimization ability of the back propagation neural network. Then, the software reliability growth model of WAGA-BPNN is presented, and the experimental data of the software reliability improvement process of quantum DC electricity meter is modeled and verified. Experiments show that the prediction accuracy of the model doubles and meets the practical requirements.