Multi-unmanned　surface　vessel　(USV)　cooperative　hunting　is　widely　used　in　dynamic　target　hunting　in　the　presence　of　obstacles,　due　to　its　high　efficiency.　This　paper　proposes　a　cooperative　hunting　method　for　multi-USVs　based　on　trajectory　prediction　by　obstacle　relation　long　short-term　memory　(ORLSTM).　First,　the　historical　target　trajectory　is　preprocessed　by　univariate　nonlinear　regression,　and　the　obstacle　information　is　input　to　OR-LSTM　to　predict　the　trajectory　of　the　target,　with　improved　accuracy　and　convergence.　A　bionic-based　cooperative　hunting　strategy　is　proposed,　by　which　multi-USVs　ambush　the　predicted　trajectory　of　the　target　by　simulating　the　hunting　strategy　of　lions.　When　the　target　is　close　to　the　ambush　position,　the　multi-USVs　start　to　hunt　it.　Two　groups　of　simulation　experiments　were　implemented　on　the　Unity　3D　platform　to　validate　the　proposed　algorithm,　which　had　higher　accuracy　and　faster　convergence　in　target　trajectory　prediction　than　the　traditional　algorithm,　and　higher　efficiency　in　target　hunting.　IEEE