The groundbreaking advancements in artificial intelligence technology have opened up novel technical avenues for the research of projectile correction technology. In projectile correction technology, the fixed-fin roll-angle command is contingent upon the guidance azimuth command, and a strong nonlinear relationship exists between them. Conventionally, the approach has been to approximate the aerodynamic model through linearization to establish the mapping relationship between the two. However, this approach suffers from limitations such as poor engineering adaptability. By leveraging the remarkable ability of the Long Short-Term Memory (LSTM) network to model nonlinear time-series data, this study constructs an accurate mapping model between the two parameters. This model not only comprehensively accounts for the intricate nonlinear relationship but also takes into consideration factors such as inaccuracies in the initial projectile-launch information (including altitude, azimuth angle of fire, etc.). Simulation results indicate that the Circular Error Probable (CEP) of the projectile's impact point can be decreased from 33.10 m to 22.07 m.