Layered Decision Architecture for Formation Air Defense with Intention Recognition and Reinforcement Learning

This is the authors' abstract. We don't add key points for this paper.

Addressing the complexity of decision-making for large-scale formation air defense tasks with multi-objective coordination, difficulties in algorithm convergence, and poor stability, a layered decision architecture for formation air defense with intention recognition and reinforcement learning is proposed. Incorporating the idea of task decomposition, the GRU algorithm is employed to discern the hidden features of situational dynamics and temporality, thereby constructing an intention recognition model to identify enemy offensive intentions. Subsequently, the exploration and learning capabilities of deep reinforcement learning algorithms are leveraged to construct a target firepower allocation model, which facilitates agent behavior decisions. Simulation results obtained from the simulation platform demonstrate that the proposed hierarchical decision model for formation air defense exhibits superior stability and adaptability in decision-making tasks in complex scenarios.