| 引用本文: | 姚煜航,欧 铭,李 敏,等.基于机器学习的淮河流域水文干旱
预测模型的构建及其应用[J].灌溉排水学报,2025,44(12):137-147. |
| YAO Yuhang,OU Ming,LI Min,et al.基于机器学习的淮河流域水文干旱
预测模型的构建及其应用[J].灌溉排水学报,2025,44(12):137-147. |
|
| 摘要: |
| 【目的】构建并验证一种基于Boruta-CNN-LSTM混合模型的淮河流域水文干旱预测方法，为流域水资源管理和防灾减灾提供科学依据。【方法】以淮河流域为例，采用1960—2020年的逐月降水量、蒸散发和土壤湿度等数据，构建了基于Boruta-random Forest Hybridizer特征选择算法、卷积神经网络（Convolutional Neural Networks，CNN）和长短期记忆网络（Long Short-Term Memory，LSTM）的混合模型（Boruta-CNN-LSTM）。通过混合模型对淮河流域月尺度的标准化径流指数SRI进行预测，并对LSTM、Boruta-LSTM、Boruta-CNN-LSTM模型的预测结果进行对比和评估。【结果】①Boruta算法显著降低了冗余特征干扰，在所选的31个影响因子中，SPI-1、蒸散发和土壤温度等因子对模型的预测能力影响显著，其中，SPI-1在各个区域中对模型的影响最为显著。②在建立的3个模型中，Boruta-CNN-LSTM模型预测效果最优，其中R2值为0.853 6，MAE和RMSE值分别为0.262和0.352，其次是Boruta-LSTM模型。【结论】通过将LSTM模型、Boruta特征选择和CNN神经网络逐步结合，Boruta-CNN-LSTM模型能够显著提高模型的预测精度，较好的适用于淮河流域的水文干旱预测。 |
| 关键词: 水文干旱预测；Boruta特征选择；LSTM；淮河流域 |
| DOI：10.13522/j.cnki.ggps.2025144 |
| 分类号: |
| 基金项目: |
|
| A machine leaning model for hydrological drought prediction: Model development and application |
|
YAO Yuhang, OU Ming, LI Min, FENG Zilong
|
|
1. College of Water Resources Science and Engineering, Yangzhou University, Yangzhou 225000, China;
2. Jilin Province Water Resource and Hydropower Consultative Company, Changchun 130012, China
|
| Abstract: |
| 【Objective】Accurately forecasting hydrological drought is critical for proactively mitigating the detrimental impacts of water scarcity. This study proposes a model to predict hydrological drought at catchment scale. 【Method】A hybrid Boruta-CNN-LSTM model was developed, which integrates the Boruta-random forest feature selection algorithm, convolutional neural networks (CNN), and long short-term memory (LSTM) networks. The accuracy of the model was tested against monthly precipitation, evapotranspiration, and soil moisture data measured from 1960—2020 in the Huai River Basin. The accuracy of the LSTM, Boruta-LSTM and Boruta-CNN-LSTM was evaluated based on their root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2) for predicting droughts in the Huai River Basin.【Result】① The Boruta algorithm effectively reduces interference from redundant features. Among the 31 initial influencing factors selected in the models, monthly standardized precipitation index, evapotranspiration, and soil temperature affects predictive ability of the models the most in all sub-regions in the basin. ② The Boruta-CNN-LSTM model is most accurate, with R2=0.853 6, MAE=0.262 and RMSE=0.352. 【Conclusion】The hybrid Boruta-CNN-LSTM model is the most accurate for predicting spatiotemporal variation of hydrological droughts in the Huaihe River Basin. Integrating LSTM, Boruta feature selection, and CNN provides a robust model for predicting drought under different hydrological scenarios. |
| Key words: hydrological drought prediction; boruta feature selection; LSTM; Huaihe River Basin |
