Scholarlink

<p>Water quality prediction is very important in ensuring water resource security and supporting sustainable ecological development. However, since the dynamic nature and complexity of data, traditional forecasting models often fail to achieve high prediction accuracy. Therefore, a hybrid SA-CNN-BiLSTM model is proposed, which combines Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory Networks (BiLSTM), and self-attention mechanism to predict future water quality trends. This paper systematically analyzes the influence of data preprocessing and the forecasting horizon on model performance. Furthermore, the model makes a comparison to Long Short-Term Memory (LSTM), BiLSTM, CNN-BiLSTM, Support Vector Regression (SVR), Light Gradient Boosting Machine (LightGBM), and its application to other river basins is also explored. The results demonstrate the model's high accuracy in DOX and other predictions, the evaluation results showed root mean square error (RMSE) = 0.4, mean absolute error (MAE) = 0.294, mean square error (MSE) = 0.162, with the coefficient of determination (R<sup>2</sup>) reaching 0.955. The model also shows optimal performance in predicting TN, NH<sub>3</sub>N, and PH, maintaining stability across other watersheds and exhibiting strong robustness, adaptability, and generalization capabilities. Compared to traditional models, this model more efficiently extracts deeper features and utilizes an attention mechanism to enhance the weighting of key time steps and features, thereby reducing prediction errors. It provides a novel approach and technical framework for water quality prediction, offering a scientific basis for the sustainable and green development of aquatic ecosystems.</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

Water quality prediction of Pohe River reservoir based on SA-CNN-BiLSTM model

  • 期刊论文

  • 发表日期：

    2025年5月21日

查看全文
  • Qingqing Tian,
  • Qiongyao Wang,
  • Lei Guo

摘要

Water quality prediction is very important in ensuring water resource security and supporting sustainable ecological development. However, since the dynamic nature and complexity of data, traditional forecasting models often fail to achieve high prediction accuracy. Therefore, a hybrid SA-CNN-BiLSTM model is proposed, which combines Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory Networks (BiLSTM), and self-attention mechanism to predict future water quality trends. This paper systematically analyzes the influence of data preprocessing and the forecasting horizon on model performance. Furthermore, the model makes a comparison to Long Short-Term Memory (LSTM), BiLSTM, CNN-BiLSTM, Support Vector Regression (SVR), Light Gradient Boosting Machine (LightGBM), and its application to other river basins is also explored. The results demonstrate the model's high accuracy in DOX and other predictions, the evaluation results showed root mean square error (RMSE) = 0.4, mean absolute error (MAE) = 0.294, mean square error (MSE) = 0.162, with the coefficient of determination (R2) reaching 0.955. The model also shows optimal performance in predicting TN, NH3N, and PH, maintaining stability across other watersheds and exhibiting strong robustness, adaptability, and generalization capabilities. Compared to traditional models, this model more efficiently extracts deeper features and utilizes an attention mechanism to enhance the weighting of key time steps and features, thereby reducing prediction errors. It provides a novel approach and technical framework for water quality prediction, offering a scientific basis for the sustainable and green development of aquatic ecosystems.