| 117 | 0 | 194 |
| 下载次数 | 被引频次 | 阅读次数 |
【目的】随着全球变暖加剧,极端降水事件频发,内涝风险评估已成为城市防灾减灾的迫切需求,现有研究多依赖于高密度的水文监测数据,相关评估工作在数据匮乏地区难以有效开展。为此,研究以北方城市哈尔滨为研究区,从动态时序变化与静态空间两个维度实现城市内涝风险的预测与评估。【方法】 在数据匮乏的背景下以哈尔滨市为研究区,系统对比了LSTM、GRU、NBEATS及TFT四种时序模型的性能,以探究其在降水时序变化预测中的适用性及准确性。在此基础上选取综合性能最优的时序模型结合CNN模型构建两阶段评估框架:首先,通过时序模型预测未来单日降水量,并对预测的单点降水数据进行插值处理形成未来单日降水的空间分布格局;其次,将预测结果与地面高程和土地利用数据融合为网格数据,以水务集团发布的易涝点位为分类标签,通过CNN模型预测研究区日值的内涝风险等级空间分布图。【结果】结果表明,在对比的四种时序模型中,GRU模型综合表现最优,其均方根误差为0.045,平均绝对误差为0.019,决定系数达0.948。基于GRU构建的GRU-CNN两阶段模型,其内涝风险分类准确率达0.926,F1分数为0.926,决定系数为0.851,模型两个阶段的预测总用时约15min。【结论】GRU-CNN两阶段模型能够在数据匮乏条件下有效预测城市内涝风险的空间分布格局,实现了动态降水时序预测与静态空间风险评估的串联,可为数据稀缺的北方寒地城市内涝防治工作提供低成本、高效率的技术支撑与决策参考。
Abstract:[Objective] As global warming intensifies and extreme precipitation events occur more frequently, urban waterlogging risk assessment has become an urgent need for urban disaster prevention and mitigation. Existing studies mostly rely on high-density hydrological monitoring data, making assessment work difficult to effectively conduct in data-scarce regions. To address this issue, Harbin, a northern city, is selected as the study area, and the prediction and assessment of urban waterlogging risk are conducted from the two dimensions of dynamic temporal changes and static spatial patterns. [Methods] Against the background of data scarcity, the performance of four time-series models, namely LSTM, GRU, NBEATS, and TFT, was systematically compared in Harbin to explore their applicability and accuracy in predicting temporal changes in precipitation. On this basis, the best-performing time-series model was selected and combined with a CNN model to construct a two-stage assessment framework. First, the future daily precipitation was predicted using the time-series model, and the predicted point precipitation data were interpolated to form a spatial distribution pattern of future daily precipitation. Second, the predicted results were fused with digital elevation model (DEM) and land use data to generate grid data. Using the waterlogging point data released by the Water Group as classification labels, the daily spatial distribution map of waterlogging risk levels for the study area was predicted by the CNN model. [Results] The results showed that among the four compared time-series models, the GRU model achieved the best overall performance, with a root mean square error of 0.045, a mean absolute error of 0.019, and a coefficient of determination of 0.948. For the GRU-CNN two-stage model constructed based on GRU, the waterlogging risk classification accuracy reached 0.926, the F1 score was 0.926, the coefficient of determination was 0.851, and the total time consumed by the two-stage prediction process was approximately 15 minutes. [Conclusion] The GRU-CNN two-stage model can effectively predict the spatial distribution pattern of urban waterlogging risk under data-scarce conditions, achieving the integration of dynamic precipitation time-series prediction and static spatial risk assessment. It provides low-cost, high-efficiency technical support and decision-making reference for urban waterlogging prevention and control in data-scarce cold-region cities of northern China.
[1] 王会军.全球变暖已经导致降水变率和灾害风险加剧了吗?[J].中国科学:地球科学,2025,55(2):662-664.
[2] ZHANG W, ZHOU T, WU P. Anthropogenic amplification of precipitation variability over the past century[J]. Science, 2024, 385: 427-432.
[3] PENDERGRASS A G, KNUTTI R, LEHNER F, et al. Precipitation variability increases in a warmer climate[J]. Scientific Reports, 2017, 7: 17966.
[4] 王子源,蔡敏,廖卫红,等. 考虑雨水口堵塞及处置效果的城市内涝模拟[J]. 南水北调与水利科技(中英文),2025,23(3):598-608.
[5] 黄楚雯,管永乐,王洪发.基于RF-LSTM模型的城市洪涝积水模拟研究[J].人民黄河,2025,47(6):50-56.
[6] 张挺,马睿佳,程泳铭,等. 基于 VMD-LSTM 模型的流域径流预报技术[J]. 南水北调与水利科技(中英文),2025,23(5):1196-1203.
[7] 胡昊,陈军朋,李擎,等. 基于CNN-GRU-ATT的城市暴雨积水预测研究[J]. 华北水利水电大学学报(自然科学版), 2024, 45(4): 27-35.
[8] 代鑫,黄弘,汲欣愉,等.基于机器学习的城市暴雨内涝时空快速预测模型[J].清华大学学报(自然科学版),2023,63(6):865-873.
[9] 齐旭,芦钰冰.基于LSTM-TCN模型的尚庄炉水库中长期径流预报研究[J].水利科学与寒区工程,2025,8(7):30-34.
[10] HABIB A B, ALAM M G R, UDDIN M Z. AUNET (attention-based unified network): Leveraging attention-based N-BEATS for enhanced univariate time series forecasting[J]. IEEE Access, 2025, 13: 95184-95217.
[11] VAN BELLE J, CREVITS R, CALJON D, et al. Probabilistic forecasting with modified N-BEATS networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2024, 35(12): 18872-18885.
[12] 吴彬溶,王林.基于JADE-TFT模型的可解释性旅游需求预测研究[J].运筹与管理,2024,33(8):148-154.
[13] 单晶,薛丰昌,程瑶.城市内涝ChebNet-LSTM积水预测模型[J].武汉大学学报(信息科学版),2025,50(6):1225-1234.
[14] 于小雅.基于深度学习的洪涝场景识别及风险评估方法研究[D].武汉:华中科技大学,2024.
[15] ANIK M S B M, AN C, LI S S. Evolution from the physical process-based approaches to machine learning approaches to predicting urban floods: A literature review[J]. Environmental Systems Research, 2025, 14(1): 15.
[16] KARIM F, ARMIN M A, AHMEDT-ARISTIZABAL D, et al. A review of hydrodynamic and machine learning approaches for flood inundation modeling[J]. Water, 2023, 15(3): 566.
[17] 寇双, 司振江. 哈尔滨市水生态与城市化耦合协调发展研究[J]. 水利科学与寒区工程, 2025, 8(9): 77-81.
[18] 胡龙颂,曾建玲,王艳艳,等.降雨量等级与空间分布不均匀性对降雨空间插值方法精度的影响[J].水资源与水工程学报,2025,36(3):73-79.
[19] 张海滨,蒋巍巍,唐云武,等.基于优化随机森林的散热器弯管壁厚时间序列预测模型[J].大电机技术,2025,(1):117-123.
[20] 钟琦,梁红丽,代刊,等.降水预报机器学习订正及其在分类型降水测试的效果研究[J].气象,2025,51(8):914-927.
[21] YANG J, HUANG X. The 30 m annual land cover dataset and its dynamics in China from 1990 to 2019[J]. Earth System Science Data, 2021, 13(8): 3907-3925.
[22] HAN Z, ZHAO J, LEUNG H, et al. A review of deep learning models for time series prediction[J]. IEEE Sensors Journal, 2021, 21(6): 7833-7848.
[23] LIM B, ZOHREN S. Time-series forecasting with deep learning: A survey[J]. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 2021, 379(2194): 20200209.
[24] TORRES J F, HADJOUT D, SEBAA A, et al. Deep learning for time series forecasting: A survey[J]. Big Data, 2021, 9(1): 3-21.
[25] ZHU J, JIANG Q S, SHEN Y H, et al. Application of recurrent neural network to mechanical fault diagnosis: A review[J]. Journal of Mechanical Science and Technology, 2022, 36(2): 527-542.
[26] OLIVARES K G, CHALLU C, MARCJASZ G, et al. Neural basis expansion analysis with exogenous variables: Forecasting electricity prices with NBEATSx[J]. International Journal of Forecasting, 2023, 39(2): 884-900.
[27] FAN K L, YANG C, XIE Z L, et al. Load-regulation characteristics of gas turbine combined cycle power system controlled with compressor inlet air heating[J]. Applied Thermal Engineering, 2021, 196: 117285.
[28] 庞楦弋.基于多模态数据与LSTM模型的大豆期货价格预测[D].重庆:重庆理工大学,2025.
[29] 贺晨煜,陆建荣.基于LSTM-GRU的水质监控系统设计与预测方法研究[J].物联网技术,2025,15(24):35-38.
[30] ABUMOHSEN M, OWDA A Y, OWDA M. Electrical load forecasting using LSTM, GRU, and RNN algorithms[J]. Energies, 2023, 16(5): 2283.
[31] 赵文路.基于集成GRU神经网络的电力系统短期负荷区间预测方法研究[D].沈阳:沈阳师范大学,2024.
[32] 马萍.基于元学习框架的小样本时间序列预测方法研究[D].杭州:浙江工商大学,2024.
[33] 罗雯丽. 基于深度学习的PM_(2.5)浓度时空特征分析及预测模型研究[D]. 包头: 内蒙古科技大学, 2024.
[34] 郑冰燕.基于CNN-SVM的交通事故严重程度分析及预测[D].沈阳:沈阳建筑大学,2025.
[35] MANZOOR M, UMER M, SADIQ S, et al. RFCNN: Traffic accident severity prediction based on decision level fusion of machine and deep learning model[J]. IEEE Access, 2021, 9: 128359-128371.
[36] 李亚茹,张宇来,王佳晨.面向超参数估计的贝叶斯优化方法综述[J].计算机科学,2022,49(S1):86-92.
[37] SNOEK J, LAROCHELLE H, ADAMS R P. Practical Bayesian optimization of machine learning algorithms[J]. arXiv preprint arXiv:1206.2944, 2012.
[38] LAN C F, SONG B W, ZHANG L, et al. State prediction of hydro-turbine based on WOA-RF-adaboost[J]. Energy Reports, 2022, 8: 13129-13137.
[39] 赵成兵,刘丹秀,谢新平,等.基于时间序列的季节性气温预测研究[J].安徽建筑大学学报,2022,30(3):83-89.
[40] BONAH E, HUANG X Y, YI R, et al. Vis-NIR hyperspectral imaging for the classification of bacterial foodborne pathogens based on pixel-wise analysis and a novel CARS-PSO-SVM model[J]. Infrared Physics and Technology, 2020, 105: 103220.
[41] 胡山,朱向华.基于ResNet18与K折交叉验证的电机轴承故障诊断[J].自动化与仪表,2025,40(4):40-45.
基本信息:
中图分类号:TP18;TU992
引用信息:
[1]刘宇烁,李苗,吴京晏,等.数据匮乏背景下基于时空深度学习算法的城市内涝风险评估:以哈尔滨为例[J].水利水电技术(中英文)().
基金信息:
国家自然科学基金项目(42501441); 新时代龙江优秀硕博论文项目(LJYXL2024-024); 黑龙江省博士后基金(LBH-Z25199,LBH-Z24207)
2026-07-31
2026-07-31
2026-07-31