In　previous　studies,　the　general　method　for　flight　recovery　problem　used　fixed　flight　transit　time,　rather　than　considered　the　result　of　flight　transit　time　changes　in　real　airports.　We　propose　a　LightGBM　model　to　predict　accurate　transit　time　based　on　the　airport-flight　features　from　total　235　airports　and　all　flights　in　China.　The　numerical　results　show　that　our　model　has　6.783　minutes　root　mean　square　error　using　real　flights　data.　We　construct　an　irregular　flight　recovery　model　based　on　effective　transit　time,　and　specifically　design　a　column　vector　generation　algorithm　to　solve　this　model.　This　algorithm　can　solve　the　problem　of　airport　traffic　flow　decrease,　airport　closure,　aircraft　maintenance　and　other　irregular　conditions　under　the　goal　of　minimizing　flight　delays,　the　number　of　cancellations,　and　the　number　of　aircraft　changes　by　canceling,　changing　the　planned　time,　and　replacing　aircraft.　Tests　on　actual　operating　data　of　airlines　prove　that　the　irregular　flight　recovery　method　based　on　transit　time　prediction　is　effective.　The　real　case　of　large-scale　flight　delays　test　shows　the　total　delay　time　can　be　reduced　by　34.2%.　The　comparison　between　the　spatio-temporal　network　algorithm　and　the　column　vector　generation　algorithm　shows　that　the　proposed　flight　recovery　method　also　can　reduce　the　recovery　cost　under　the　premise　of　the　same　recovery　result.　©　2022,　Editorial　Board　of　JBUAA.　All　right　reserved.