Abstract: Numerical simulation of liquid sloshing force under lateral acceleration of reference tank was conducted with volume of fluid (VOF) model of fluent software. Liquid filling ratio, forces, acceleration were set as inputs, while forces after a few seconds as target outputs. Reasonable distributed results were selected as training samples to establish prediction model of liquid sloshing forces acting on the reference tank. This trained BP neural networks were tested by 158 various samples. The maximum prediction errors of vertical force, lateral force and lateral moment were 8.88%, 0.36%, and 1.38%. The magnitude and regulation of the modified time step and forces of reference tank are nearly the same as general cylindrical tanks; The magnitudes of forces are proportional to the length of cylindrical tank in lateral motion. Through revising the time step and forces of reference tank, the forces of general cylindrical tank can be predicted based on BP neural network. The study can provide the necessary data for the analysis of tank truck dynamics efficiently.