A multi-agent jamming decision-making method based on SP-HASAC
Jingpeng Gao, Chen Shen, Lu Gao, Zhiye Jiang · IET conference proceedings. · 2024
In the study of the multi-UAV penetration mission decision method, the Heterogeneous Agent Soft Actor-Critic algorithm uses the HAML framework and the maximum entropy principle to enable UAVs to have stronger anti-jamming capability and exploration capability. However, it is difficult to accurately evaluate the action value of UAVs and the training stability is poor. To address this issue, this paper proposes a multi-agent based on the state prediction-HASAC (SP-HASAC) jamming decision method. Based on the idea of advantage function decomposition and environmental dynamics model, the state prediction network is designed to extract more abundant spatial information of penetration battlefield and achieve accurate evaluation of the action value of multi-UAV. Combining with the negative domain activation theory, the ReLU function is improved to realize the stable training of multi-UAV. The simulation results show that in the electronic penetration scenario of three UAVs against a single radar, the training convergence time of the proposed improved algorithm is reduced by 22% compared with the original algorithm. Three UAVs penetration to twice the burn-through distance still has more than 70% penetration success rate.