Extensive theoretical research has been conducted on jet head velocity in previous studies. However, due to the influence of numerous nonlinear factors, these theories cannot accurately calculate the jet head velocity under different warhead structures. To address this issue, neural networks have inherent advantages. In this paper, the charge radius r, the charge height h, the liner angle α and the liner thickness t are employed as the input for the BP neural network, while the jet head velocity serves as the output. A dataset comprising 168 sets of data is prepared for training the BP neural network, with a split ratio of 80% for training and 20% for validation. The training results demonstrate that the back propagation (BP) neural network is capable of accurately predicting the jet head velocity. The jet head velocity v increases with the increase of the charge radius r, increases with the increase of the charge height h, decreases with the increase of the liner angle α, and decreases with the increase of the liner thickness t. Compared to BP Neural Networks, RNN performs very well on datasets average error can be up to 11.63 m/s but performs poorly when dealing with new time series data. These conclusions can provide guidance for further jet design and optimization, and help to understand the relationship between jet head velocity and warhead structure.