With　the　introduction　of　the　concept　of　＇Intelligent　Transport＇,　in　order　to　meet　the　needs　of　high　accuracy　of　flight　delay　prediction　in　the　transportation　industry　in　the　current　stage,　a　short-term　flight　delay　prediction　model　based　on　multi-model　fusion　is　proposed　in　this　paper.　The　data　used　in　this　study　come　from　the　official　data　of　Ctrip.　First　of　all,　the　model　in　this　paper　extracts　and　normalizes　the　urban　weather　information,　time　information,　airline　information,　delay　information　and　average　information,　and　solves　the　over-fitting　problem.　Then,　add　L2　regularization　to　solve　the　over-fitting　problem.　Finally,　the　fusion　prediction　model　proposed　in　this　paper　is　used　to　cross-verify　and　predict　the　integrated　data.　The　structure　of　the　fusion　model　based　on　stacking　proposed　in　this　paper　extracts　the　advantages　of　both　random　forest,　XGBoost　and　LightGBM,　constructs　it　by　using　the　method　of　cross-validation,　and　combines　multiple　models　to　judge　the　results　for　secondary　training.　This　method　can　not　only　enhance　robustness　but　also　better　resist　over-fitting.　According　to　the　experimental　results,　the　prediction　accuracy　and　stability　of　flight　delay　are　improved　by　using　this　model.　©　2020　IEEE.