Multi-Step State Forecasting with Conformal Prediction for High-Rate Dynamic Systems

Abstract

High-rate systems, such as hypersonic vehicles and impact mitigation mechanisms, exhibit rapid, nonlinear, and rapid time-varying dynamics. These systems are often subjected to extreme accelerations (e.g., > 100 gn) over very short durations (e.g., < 100 ms), making them highly susceptible to uncertainties, non-stationarities, and external disturbances. In such environments, accurate and reliable state forecasting is essential for real-time decision-making and control. This paper presents a novel framework for high-rate multi-step forecasting that integrates recurrent neural networks (RNNs), topological data analysis (TDA), and conformal prediction for uncertainty quantification. RNNs are used to model temporal dependencies in high-dimensional system dynamics, while TDA-derived features capture the underlying topological structure of the system state. To provide robust and interpretable uncertainty estimates, we employ conformal prediction techniques—particularly inductive and block-based conformal methods— to construct statistically valid prediction intervals. These methods adapt to the non-exchangeable nature of time-series data, enabling coverage guarantees under temporal dependence. The effectiveness of the proposed framework is evaluated on experimental data from the Dynamic Reproduction of Projectiles in Ballistic Environments for Advanced Research (DROPBEAR) testbed. Results demonstrate that the model produces accurate multi-step forecasts with well-calibrated uncertainty bounds, which can be used to support high-rate system monitoring and decision-making.