In　this　paper,　we　mainly　study　the　application　of　Long　Short-Term　Memory　(LSTM)　algorithms　in　the　stock　market.　LSTM　originates　from　the　recurrent　neural　network　(RNN)　and　has　a　significant　effect　on　the　time　series　problems.　In　this　paper,　the　BP　neural　network　model　and　the　LSTM　model　are　established　respectively.　Then　we　combine　them　with　the　stock　data,　a　series　of　prediction　results　are　obtained.　Obviously,　the　prediction　results　of　LSTM　model　are　more　accurate,　and　the　prediction　accuracy　rate　can　reach　60%-65%.　In　the　modeling　process,　in　order　to　solve　the　＇saw-tooth　phenomenon＇　of　the　gradient　descent　algorithm　which　is　inevitable,　we　have　improved　the　traditional　gradient　descent　algorithm　and　specially　designed　the　input　data　of　the　neural　network.　In　addition,　we　defined　a　parameter　combination　library　and　use　the　skill　of　dropout　to　get　the　more　ideal　prediction　results.　©　2018　IEEE.