考虑高峰偏差的城轨新线车站分类及影响机制OA

A scientifically grounded classification of stations on new urban rail transit lines is essential for identifying station functions and anticipating passenger flow characteristics,thereby supporting optimized planning and design and improved operational efficiency.However,existing studies have focused primarily on stations already in operation.Since the corresponding methods generally rely on historical passenger flow data,they cannot be directly applied during the planning stage of new lines.Moreover,the mismatch between station-level and line-level peak periods has generally been overlooked,causing classification results to deviate from actual passenger flow characteristics.To address these limitations,this study focused on new-line planning scenarios and,for the first time,incorporated peak-period deviation into the station classification system.A nonlinear and interpretable machine-learning framework integrating K-means++,SMOTE,XGBoost,and SHAP was proposed.First,a feature system comprising peak deviation time,peak deviation coefficient,and weekend scale coefficient was constructed.Nonnegative tensor factorization was then employed to extract latent station factors,which were combined with these indicators to form a composite feature set.The K-means++algorithm was subsequently used to cluster existing stations and identify typical peak-period passenger flow patterns.Second,to address class imbalance,a SMOTE-XGBoost model was developed to learn the complex nonlinear mapping between built-environment features and station classes.Based on this mapping,the class and peak-period passenger flow pattern of a station on a new line can be predicted during the planning stage using its built-environment data.Furthermore,SHAP-based attribution analysis was introduced to identify the dominant factors affecting different station classes and reveal their underlying mechanisms.The results show that:(1)the proposed method can accurately predict station classes before a new line opens,and the predicted classes are consistent with the actual operational characteristics.The agreement rate with the benchmark method reaches 94%,which is 51 percentage points higher than that of the ablation method excluding peak-period deviation(43%).The method also performs well across datasets from multiple lines,demonstrating its effectiveness and broad applicability;and(2)different station classes are influenced by different dominant factors.Nevertheless,the number of bus routes and the distance to the city center are fundamental explanatory variables across all station classes.The former exhibits class-specific threshold effects,whereas the latter shows multistage threshold effects that vary with spatial distance.The proposed model can provide forward-looking decision support for the planning,design,and operational preparation of stations on new urban rail transit lines and can also accommodate station class updates in response to dynamic changes in the built environment.