This　paper　analyzes　the　time-varying　patterns　of　urban　bus　passenger　flow　and　predicts　future　short-term　bus　passenger　flow,　which　helps　public　transport　managers　to　predict　bus　passenger　flow　in　advance　and　adjust　bus　scheduling　plans.　This　paper　constructs　a　short-term　passenger　flow　forecasting　method　based　on　the　Gradient　Boosting　Decision　Tree　(GBDT)　model　and　introduces　the　three-structured　Parzen　Estimator　Approach　(TPE)　to　optimize　the　parameter　space.　Results　showed　that　the　prediction　model　proposed　in　this　paper　can　make　full　use　of　the　multi-feature　vector　data　to　predict　the　various　passenger　flow　pattern　and　has　a　lower　prediction　error　compared　with　the　GBDT　base　model　and　other　models.　The　model　can　further　improve　the　accuracy　of　short-time　passenger　flow　prediction　and　provide　important　quantitative　data　support　for　bus　operation　guarantee　and　transport　scheduling　plan　optimization.　©　ASCE.