Online Parameter Identification for an Incoming Interceptor with Observability Analysis

Reliable identification of the key parameters governing an incoming interceptor's trajectory is essential for accurate maneuver prediction and effective defensive guidance in air combat. However, these parameters are difficult to obtain directly due to weak observability and nonlinear coupling within the system dynamics, while traditional approaches struggle to achieve fast and accurate regression-based identification of multiple parameters. To address this challenge, a neural network (NN)-based identification framework is proposed. Firstly, an observability analysis is conducted to examine the fundamental identifiability of the interceptor's parameters. Based on this analysis, a Gated Recurrent Unit (GRU) neural network integrated with a cross operation and an improved multiple-model mechanism (Cross-GRU-IMMM) is developed. The cross operation enhances feature correlations, while the GRU layers and multi-head self-attention jointly capture and refine temporal dependencies. A multiple-model layer then aggregates the weighted outputs through a Softmax-based mechanism to produce the final parameter estimates. Furthermore, the analytical derivation of the back-propagation process in the multiple model layer provides guidance for efficient parameter optimization. Simulation results demonstrate that the proposed method achieves significantly faster convergence speed and higher identification accuracy compared with conventional Cubature Kalman Filter (CKF), CKF-based Multiple-Model Adaptive Estimation (MMAE) and NN-based approaches. Ablation experiments further confirm the complementary roles of these components in enhancing performance. And a comprehensive sensitivity analysis is also conducted to investigate the influence of noise and input steps. These results verify the proposed framework's effectiveness and scalability for real-time parameter identification in uncertain engagement environments.