UAV countermeasure maneuver decision based on deep reinforcement learning

The struggle for air supremacy is the core issue in modern military confrontation. In recent years, deep reinforcement learning has developed rapidly in the field of decision-making. Compared with traditional reinforcement learning methods, deep reinforcement learning can effectively solve continuous decisionmaking problems in large-scale state spaces. Based on the theory of deep reinforcement learning, this paper conducts research on UAV confrontation decision-making, combined with neural network to fit the action value function of the large-scale state space in the aerial combat problem, and implements it based on the exploration strategy combining E-value and $\epsilon$-greedy reinforce the balance of exploration and utilization in the learning process, and conduct a comparative experiment. Compared with the DQN maneuver decision algorithm based on the attenuated greedy algorithm, it improves the convergence speed of the aerial combat algorithm by 100% and greatly reduces the loss function.