Missile Trajectory Prediction Under Diverse Error Conditions Using Hypernetwork-Based Continual Learning

Missile trajectory prediction under various error conditions is critical for effective mission planning and real-time operational decision-making. Previous research has typically emphasized analyzing how uncertainties influence the final impact point distribution using probabilistic methods and Monte Carlo simulations. However, these studies often overlook the dynamic evolution of individual error factors and their impact on trajectory shape throughout the flight. To fill this gap, we investigate missile trajectory prediction under diverse error conditions using a method that combines Neural Ordinary Differential Equations (Neural ODEs) with a hypernetwork-based continual learning framework. Our approach is designed not only to predict missile trajectories accurately but also to reveal how distinct error conditions dynamically shape trajectories. The hypernetwork adapts the Neural ODE to each new error scenario, while the continual learning approach allows the model to adapt efficiently to new error types preventing catastrophic forgetting of knowledge learned from earlier error scenarios. Comprehensive numerical experiments demonstrate that our model successfully captures error-specific trajectory characteristics and maintains robust predictive performance under various error conditions.