Efficient Air Defense Temporal Decision-Making Methods Under Unstable Dimensional States

To address the needs of future intelligent war-fare, this paper proposes a hierarchical intelligent decision-making algorithm (ATT-PPO) based on the PPO algorithm and multi-head attention mechanism, designed to solve air defense sequential decision-making problems under unsta-ble dimensional states. Traditional reinforcement learning struggles to handle variable-dimensional observation information, whereas ATT-PPO uses the attention mechanism to map variable-dimensional observations into fixed dimensions, thereby enhancing the representation of high-dimensional dynamic data. However, high-dimensional decision spaces increase the learning burden of the algorithm. To mitigate this, a hierarchical reinforcement learning architecture is designed, incorporating domain knowledge to narrow the exploration space and improve decision-making efficiency. Simulation results show that ATT-PPO significantly out-performs traditional expert control methods in cumulative rewards, validating its superior performance.