The　hunting　of　dynamic　targets　by　multi　unmanned　surface　vehicle(USV)　is　an　important　problem　in　USV　swarm　operations.　In　this　paper,　aiming　at　the　dynamic　target　oriented　swarm　hunting　problem,　by　analyzing　the　shortcomings　of　the　hunting　mechanism　based　on　MADDPG　algorithm,　we　introduce　the　attention　mechanism　into　the　hunting　process,　and　then　design　the　cooperative　hunting　strategy　based　on　the　attention　mechanism(Att-MADDPG),　updating　the　corresponding　hunting　algorithm.　Firstly,　the　attention　module　is　added　to　critic　network　to　process　the　information　of　all　USVs　according　to　different　attention　weights.　Secondly,　the　attention　module　is　added　to　actor　network　to　promote　other　USVs　to　carry　out　cooperative　hunting.　Finally,　establish　obstacle-free　and　obstacle　simulation　environment　according　to　the　hunting　task.　The　simulation　results　show　that　the　training　stability　and　timeliness　of　Att-MADDPG　algorithm　are　significantly　improved　compared　with　MADDPG　algorithm.　After　learning,　the　USV　can　cooperate　to　make　the　swarm　emerge　more　intelligent　behavior.