The　long-term　prestress　loss　caused　by　shrinkage　and　creep　of　concrete　and　stress　relaxation　of　prestressed　tendons　has　significant　effects　on　the　sealability　and　safety　of　prestressed　concrete　cylinder　structures　such　as　nuclear　reactor　containments　and　liquified　natural　gas　(LNG)　tanks.　By　utilizing　machine　learning　(ML)　tech-niques,　this　study　aims　to　establish　an　intelligent　approach　for　the　long-term　prestress　loss　prediction　of　concrete　cylinder　structures.　Firstly,　based　on　the　Infrastructure　Technology　Institute　of　Northwestern　University　(NU-ITI)　database　of　concrete　shrinkage　and　creep　performance,　the　explicit　expressions　are　presented　for　concrete　shrinkage　and　creep　function　using　genetic　programming　(GP);　Moreover,　the　concrete　constitutive　model　is　incorporated　into　a　general　finite-element　software　package　based　on　the　ABAQUS　UMAT　platform.　Then　finite　element　analysis　(FEA)　models　are　established　and　calibrated　based　on　the　existing　long-term　prestress　loss　tests　of　prestressed　concrete　beams.　In　addition　to　the　experimental　results　in　the　literature,　the　numerical　results　of　the　FEA　model　are　also　used　to　form　the　database　of　the　long-term　prestress　losses　for　concrete　cylinder　struc-tures.　Finally,　three　prediction　models　of　long-term　prestress　loss　are　proposed　by　utilizing　the　artificial　neural　network　(ANN),　one-dimensional　convolutional　neural　network　(1D　CNN)　and　genetic　programming　(GP).　Compared　with　the　measured　results　of　nuclear　containments　in　practical　engineering,　the　ML　based　prediction　models　are　demonstrated　to　be　accurate　and　efficient　in　evaluating　the　long-term　prestress　loss　for　prestressed　concrete　cylinder　structures.