The　Bolted　joint　loosening　is　challenging　to　describe　with　an　accurate　mathematical　model.　Therefore,　a　prediction　model　of　bolted　joint　loosening　was　proposed　based　on　a　deep　learning　network.　Firstly,　the　loosening　experiments　of　the　bolted　joint　were　carried　out　with　an　orthogonal　test.　And　the　nonlinear　and　uncertain　characteristics　of　bolted　joint　loosening　process　were　further　analyzed.　Furthermore,　a　prediction　model　of　bolted　joint　loosening　based　on　a　deep　learning　network　was　established.　The　proposed　model　was　trained　and　verified　using　the　experimental　data.　The　results　showed　that　compared　with　the　traditional　mathematical　and　regression　models,　the　model　could　not　only　obtain　the　change　of　the　mean　value　of　preload　but　also　describe　the　confidence　interval　of　the　change　of　preload　in　the　sense　of　probability　synchronously.　The　experiment　data　and　prediction　data　are　in　good　agreement,　which　can　be　used　to　validate　the　rationality　of　the　model.