Efficient and robust guidance for active aircraft defense: A hybrid-screened associated globalized dual heuristic programming approach

High-dimensional nonlinearity and intense adversarial interactions in Target-Defender-Attacker ( T-D-A ) engagements pose significant challenges to active aircraft defense. To address these complexities, this paper proposes a unified active defense decision framework for online synchronous policy optimization, simultaneously generating optimal evasion maneuvers for the target aircraft and precision interception commands for the defender missile. To resolve the strong inter-agent coupling and cross-scale dynamics, a Hybrid-Screened Associated Globalized Dual Heuristic Programming (HS-AGDHP) algorithm is developed. At the architectural core is a novel associated dual network enhanced by hybrid-scale activation functions, which effectively captures heterogeneous combat dynamics and alleviates the spectral bias of traditional MLPs. Unlike conventional adaptive dynamic programming (ADP) relying on noise-sensitive automatic differentiation, this architecture explicitly outputs both the performance index and its costates in a single forward pass. By bypassing the need for recursive backpropagation, this gradient-free mechanism reduces computational overhead, demonstrating the theoretical potential for online guidance applications. To fortify this structure, a physics-heuristic screening mechanism based on Stochastic Configuration Networks (SCN) selects noise-insensitive nodes, effectively mitigating noise in stochastic environments. Furthermore, a saturation-embedded Actor network effectively bounds the physical actuators to prevent policy oscillation under aggressive maneuvers, while a Time-Varying Incremental Model (TVIM) alleviates the dependence on global prior dynamics. Comprehensive statistical simulations demonstrate that HS-AGDHP framework achieves a strong comprehensive balance among command smoothness, control effort efficiency, and environmental robustness.