Scholarlink

<p>In hydrology, runoff forecasting is essential for disaster relief, flood management, and the efficient use of water resources. To address the problem of low prediction accuracy of the traditional methods, a hybrid model named SWT-ESN-CNN-PSO is proposed. First, smooth wavelet transform (SWT) is used to decompose runoff data into multi-resolution components for trend identification to capture long-term trends and short-term fluctuations. Then, a new architecture based on echo state network (ESN) and convolutional neural network (CNN) is employed where the ESN is aimed to capture temporal dependencies of decomposed runoff series and the CNN is used to further extract the reserve pooling state of the ESN. Finally, the particle swarm optimization (PSO) algorithm is employed to optimize the key parameters of ESN. Experiments based on the runoff data from Waizhou station in the Ganjiang River basin and Ankang station in the Hanjiang River basin show that the proposed SWT-ESN-CNN-PSO model outperforms tradition models of AR, LSTM-PSO, ESN-PSO, and ESN-CNN-PSO. It achieves the best prediction accuracy, with correlation coefficients (R) of 0.9880 and 0.9718 for the two hydrological stations, respectively.</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 deep learning framework incorporating wavelet transform for monthly runoff prediction

  • 期刊论文

  • 发表日期：

    2025年9月26日

查看全文
  • Yingshuang Liu,
  • Xinqiong Liu,
  • Yirong Wu,
  • Tinglong Tang,
  • Shuifa Sun,
  • Sixian Chan,
  • Yanhua Zhu

摘要

In hydrology, runoff forecasting is essential for disaster relief, flood management, and the efficient use of water resources. To address the problem of low prediction accuracy of the traditional methods, a hybrid model named SWT-ESN-CNN-PSO is proposed. First, smooth wavelet transform (SWT) is used to decompose runoff data into multi-resolution components for trend identification to capture long-term trends and short-term fluctuations. Then, a new architecture based on echo state network (ESN) and convolutional neural network (CNN) is employed where the ESN is aimed to capture temporal dependencies of decomposed runoff series and the CNN is used to further extract the reserve pooling state of the ESN. Finally, the particle swarm optimization (PSO) algorithm is employed to optimize the key parameters of ESN. Experiments based on the runoff data from Waizhou station in the Ganjiang River basin and Ankang station in the Hanjiang River basin show that the proposed SWT-ESN-CNN-PSO model outperforms tradition models of AR, LSTM-PSO, ESN-PSO, and ESN-CNN-PSO. It achieves the best prediction accuracy, with correlation coefficients (R) of 0.9880 and 0.9718 for the two hydrological stations, respectively.