Scholarlink

<p>Accurate streamflow forecasting is essential for effective water resource management, flood control, and environmental sustainability. This study proposes a novel Spatio-Temporal Attention Federated Learning (STAF-L) model that integrates spatial and temporal attention mechanisms within a federated learning framework to enhance predictive performance while preserving data privacy. The model captures complex spatio-temporal dependencies by dynamically emphasizing relevant features during decentralized training. The proposed approach is evaluated using streamflow data from the Ala River basin and compared with baseline models, including Multi-layer Perceptron (MLP), Long Short Term Memory (LSTM), Support Vector Machines (SVM), and Convolutional Neural Networks (CNN). Performance is assessed using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Coefficient of determination (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^{2}\)</EquationSource> </InlineEquation>), Precision, Recall, F1-score, and computational runtime. Experimental results show that the STAF-L model achieves significant improvements, including RMSE reductions of 18.14%, 11.81%, 42.74%, and 24.13% over CNN, SVM, LSTM, and MLP, respectively, and MAE improvements of up to 69.55% across different basin datasets. The model also attains high predictive performance, with <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(R^{2}\)</EquationSource> </InlineEquation> values up to 0.924 and superior classification metrics, including Precision of 91 and Recall of 89. Furthermore, the federated learning framework ensures computational efficiency, achieving an optimal training time of approximately 366 milliseconds per round, while enabling robust learning from distributed datasets without compromising accuracy or data privacy.</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

Spatio-Temporal Attention in Federated Learning for Streamflow Forecasting

  • 期刊论文

  • 发表日期：

    2026年5月12日

查看全文
  • Aliyu Ashiru,
  • Hongsong Chen,
  • Suhail Ahmed Rajpar

摘要

Accurate streamflow forecasting is essential for effective water resource management, flood control, and environmental sustainability. This study proposes a novel Spatio-Temporal Attention Federated Learning (STAF-L) model that integrates spatial and temporal attention mechanisms within a federated learning framework to enhance predictive performance while preserving data privacy. The model captures complex spatio-temporal dependencies by dynamically emphasizing relevant features during decentralized training. The proposed approach is evaluated using streamflow data from the Ala River basin and compared with baseline models, including Multi-layer Perceptron (MLP), Long Short Term Memory (LSTM), Support Vector Machines (SVM), and Convolutional Neural Networks (CNN). Performance is assessed using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Coefficient of determination ( \(R^{2}\) ), Precision, Recall, F1-score, and computational runtime. Experimental results show that the STAF-L model achieves significant improvements, including RMSE reductions of 18.14%, 11.81%, 42.74%, and 24.13% over CNN, SVM, LSTM, and MLP, respectively, and MAE improvements of up to 69.55% across different basin datasets. The model also attains high predictive performance, with \(R^{2}\) values up to 0.924 and superior classification metrics, including Precision of 91 and Recall of 89. Furthermore, the federated learning framework ensures computational efficiency, achieving an optimal training time of approximately 366 milliseconds per round, while enabling robust learning from distributed datasets without compromising accuracy or data privacy.