Scholarlink

<p>Water quality prediction is challenging due to the complex temporal oscillations inherent in time series data. This study addressed these challenges by proposing SSA-optimized CEEMDAN-VMD-CNN-BiLSTM-Attention (SCV-CBA)&#xa0;hybrid model to enhance prediction accuracy. The method began by decomposing water quality time series data into Intrinsic Mode Functions (IMFs) using CEEMDAN, which captured oscillatory patterns at different time scales. SSA-optimized k-means clustering grouped the IMFs into high, medium, and low-frequency components, with high-frequency components further refined using SSA-optimized VMD for detailed feature extraction. CNN was employed to capture spatial patterns in the data, while BiLSTM network modelled temporal dependencies. Attention mechanism dynamically emphasized key features in the time series. The integrated predictions from all components delivered robust and precise results. In dissolved oxygen prediction, the model outperformed others with MSE of 0.08925, RMSE of 0.29875, MAE of 0.20300, and R<sup>2</sup> of 0.97679. Ablation test, generalization ability test, and test on varying feature dimensions demonstrated the robustness, feasibility, and potential of the model as a reliable tool for water quality prediction.</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

Research on water quality prediction of Jiangshan Port based on SCV-CBA model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月8日

查看全文
  • Yiting Xu,
  • Zhaoju Liu

摘要

Water quality prediction is challenging due to the complex temporal oscillations inherent in time series data. This study addressed these challenges by proposing SSA-optimized CEEMDAN-VMD-CNN-BiLSTM-Attention (SCV-CBA) hybrid model to enhance prediction accuracy. The method began by decomposing water quality time series data into Intrinsic Mode Functions (IMFs) using CEEMDAN, which captured oscillatory patterns at different time scales. SSA-optimized k-means clustering grouped the IMFs into high, medium, and low-frequency components, with high-frequency components further refined using SSA-optimized VMD for detailed feature extraction. CNN was employed to capture spatial patterns in the data, while BiLSTM network modelled temporal dependencies. Attention mechanism dynamically emphasized key features in the time series. The integrated predictions from all components delivered robust and precise results. In dissolved oxygen prediction, the model outperformed others with MSE of 0.08925, RMSE of 0.29875, MAE of 0.20300, and R2 of 0.97679. Ablation test, generalization ability test, and test on varying feature dimensions demonstrated the robustness, feasibility, and potential of the model as a reliable tool for water quality prediction.