Abstract: In response to the large error in high-speed signal state judgment in the fuse control system and the shortcomings of BP（back propagation, BP） neural network in the state judgment process, such as poor accuracy and easy trapping in local optimal solutions, some optimization algorithms were used to improve the shortcomings of BP neural network and reduced the error in high-speed signal state judgment. Genetic algorithm was used to optimize the BP neural network for building a model, and the high-speed signal time and voltage of the fuse were made as input indicators to establish a classification model. The model was used to judeg the high-speed signal state in order to improve recognition accuracy, accelerate convergence speed, and reduce errors. And the signal state was used to know the state of the fuse control system at each moment and judge whether the system was normal and reliable. The simulation results show that the method proposed in this article has the characteristics of excellent recognition results, fast convergence speed, and small error in the high-speed signal state judgment of the fuse. Its accuracy rate reaches 99. 6%, which is better than the 88. 6%of BP neural network and 98. 7% of convolutional neural network. At the same time, the average absolute error is reduced to 0. 012 10, the mean square error is reduced to 0. 043 68, and the root mean square error is reduced to 0. 209 01. The evolution generation is 23 generations, Better than BP neural network with 0. 168 42, 0. 319 85, 0. 564 75, and 51st generation; 0. 022 63, 0. 060 5, 0. 245 97, 25th generation of convolutional neural networks. The continuous experiments results show that the improved model has better robustness. The Wilcoxon rank sum test results also show that the improved model has better recognition performance and better generalization ability compared to BP neural network and convolutional neural network.The model meets the requirements of high-speed signal state judgment.