The survival rate of a single aircraft in the face of opponent aerial suicide drone attacks is often low during flight missions, so this paper tries to explore a way to disrupt opponent tracking targets through coordinated maneuvers to solve this problem. In this paper, we set up a 2-on-2 confrontation scenario, randomly initialize the positions of both sides, and carry out research under the condition that our aircraft’s lateral maneuverability is significantly weaker than that of the opponent. Based on the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm under the Centralized Training and Decentralized Execution (CTDE) framework, each aircraft in our camp is defined as an independent agent that makes distributed decisions through offline learning. The success condition of the mission set in this paper is that at least one single vehicle survives. After 5000 experiments, the survival rate is about 75%.