Switching-aware multi-agent deep reinforcement learning for target interception