| 引用本文: | 程国柱,吕岩峰,冯天军.灾害性天气条件下城市轨道交通客流DRN-BiLSTM预测模型[J].哈尔滨工业大学学报,2025,57(7):153.DOI:10.11918/202407003 |
| CHENG Guozhu,Lü Yanfeng,FENG Tianjun.Urban rail transit passenger flow DRN-BiLSTM combined forecasting model under catastrophic weather conditions[J].Journal of Harbin Institute of Technology,2025,57(7):153.DOI:10.11918/202407003 |
|
| 摘要: |
| 为有效应对灾害性天气对城市轨道交通客流组织的影响,探索灾害性天气条件下城市轨道交通线网客流变化规律,基于深度残差网络（DRN）和双向长短时卷积神经网络（BiLSTM）,充分考虑灾害性天气特征对客流变化的影响,开展了轨道交通线网客流预测研究。构建了融入灾害性天气特征的轨道交通线网客流DRN-BiLSTM预测模型,并选取均方误差（MSE）、均方根误差（RMSE）、平均绝对误差（MAE）、决定系数（R2）作为模型性能评价指标,并进行模型的客流量预测效果验证分析。分析结果表明:与传统LSTM、BiLSTM相比较,DRN-BiLSTM在灾害性天气特征输入情况下,MSE分别降低22.10%、21.96%；RMSE分别降低10.54%、10.46%；MAE分别降低3.20%、3.95%；R2分别提升5.01%、2.12%。使用网格搜索法对模型参数进行调优,优化后,模型训练损失降低36%。通过实例验证了所构建的轨道线网交通客流预测DRN-BiLSTM组合模型能够有效捕捉数据的深层特征,极大提升了客流预测精度。 |
| 关键词: 城市交通 客流预测模型 深度残差网络 双向长短时神经网络 灾害性天气 |
| DOI：10.11918/202407003 |
| 分类号:U491 |
| 文献标识码:A |
| 基金项目:中央高校基本科研业务费专项资金(2572023CT21)；吉林省科技发展计划项目(20220402030GH) |
|
| Urban rail transit passenger flow DRN-BiLSTM combined forecasting model under catastrophic weather conditions |
|
CHENG Guozhu1,Lü Yanfeng1,FENG Tianjun2
|
|
(1.School of Civil Engineering & Transportation, Northeast Forestry University, Harbin 150040, China; 2.School of Transportation Science & Engineering, Jilin University of Architecture, Changchun 130118, China)
|
| Abstract: |
| To effectively address the impact of catastrophic weather on urban rail transit passenger flow organization and to explore the rules of passenger flow changes under such weather conditions, this paper conducts a prediction study on urban rail transit network passenger flow. Based on a deep residual network (DRN) and bidirectional long short-term memory (BiLSTM), a DRN-BiLSTM prediction model incorporating catastrophic weather features is constructed. The model′s performance is evaluated using metrics such as mean squared error (MSE), root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2), and its passenger flow prediction effectiveness is verified and analyzed. The results show that compared to traditional LSTM and BiLSTM, when inputting catastrophic weather features, the DRN-BiLSTM model reduces MSE by 22.10% and 21.96%, RMSE by 10.54% and 10.46%, MAE by 3.20% and 3.95%, and increases R2 by 5.01% and 2.12%, respectively. By optimizing model parameters with the grid search method, the model training loss is reduced by 36%. Practical verification demonstrates that the DRN-BiLSTM combined model constructed in this paper can effectively capture deep data features and significantly improve the accuracy of passenger flow prediction. |
| Key words: urban traffic passenger flow prediction model deep residual networks bidirectional long short-term convolutional neural networks catastrophic weather conditions |
