Scholarlink

<p>Runoff simulation is a critical component of flood prevention and mitigation strategies, as well as water resource management in watersheds. This study proposes a rolling forecasting model based on Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM) and attention mechanisms. The model integrates the feature extraction capabilities of CNN, the strengths of LSTM in time series modeling, and the weighted focus of attention mechanisms to enhance the accuracy of runoff forecasting. Rolling forecasting enables the model to continuously update and adapt based on the latest data, making the forecasting process more dynamic and responsive. The model is further verified and applied in the Wei River basin, the largest tributary of the Yellow River basin of China. Results indicate that the CNN-LSTM-Attention (CLA) coupling model achieves R<sup>2</sup> greater than 0.80 and NSE greater than 0.70, demonstrating greater precision in daily runoff forecasting with respect to traditional models and effectively capturing complex features in time series data. The rolling forecasting results of CLA coupling model exhibits high accuracy and stability across various hydrological stations, with both R<sup>2</sup> and NSE exceeding 0.90. This study proposes a novel technical approach for runoff forecasting and offers a scientific basis for watershed water resource management and scheduling.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Hybrid Deep Learning Approach for Runoff Rolling Forecasting

  • 期刊论文

  • 发表日期：

    2026年5月11日

查看全文
  • Lingzi Wang,
  • Rengui Jiang,
  • Yong Zhao,
  • Jiancang Xie,
  • Fawen Li,
  • Simin Wang,
  • Ganggang Zuo

摘要

Runoff simulation is a critical component of flood prevention and mitigation strategies, as well as water resource management in watersheds. This study proposes a rolling forecasting model based on Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM) and attention mechanisms. The model integrates the feature extraction capabilities of CNN, the strengths of LSTM in time series modeling, and the weighted focus of attention mechanisms to enhance the accuracy of runoff forecasting. Rolling forecasting enables the model to continuously update and adapt based on the latest data, making the forecasting process more dynamic and responsive. The model is further verified and applied in the Wei River basin, the largest tributary of the Yellow River basin of China. Results indicate that the CNN-LSTM-Attention (CLA) coupling model achieves R2 greater than 0.80 and NSE greater than 0.70, demonstrating greater precision in daily runoff forecasting with respect to traditional models and effectively capturing complex features in time series data. The rolling forecasting results of CLA coupling model exhibits high accuracy and stability across various hydrological stations, with both R2 and NSE exceeding 0.90. This study proposes a novel technical approach for runoff forecasting and offers a scientific basis for watershed water resource management and scheduling.