Scholarlink

<p>Managing river water quality by accurately predicting Biochemical Oxygen Demand (BOD) is essential for effective pollution control, particularly in urbanised catchments such as Malaysia's Klang River. Traditional BOD₅ methods are labour-intensive and unsuitable for real-time applications. This study aims to evaluate and compare the performance of three deep learning models, namely, Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Gated Recurrent Unit (GRU), for multipoint BOD prediction using eight physicochemical characteristics collected from eleven river monitoring stations. Data were processed using interpolation (PCHIP, Makima, and the Spline), and normalisation methods (Z-score and Min–Max) were used to form uniform time series for model training and validation. The results show that BiLSTM achieved the highest predictive accuracy with an average MAE of 0.3034 and RMSE of 0.3774, GRU recorded the fastest training time (18&#xa0;s) while predictions (average MAE = 0.30 – 0.32 and RMSE = 0.37 – 0.39), GRU achieved the fastest training time (18&#xa0;s) and the lowest training-phase errors (MAE = 0.0010, RMSE = 0.0018). In contrast, LSTM yielded slightly higher errors (MAE = 0.3217, RMSE = 0.4020). These findings demonstrate that BiLSTM is most suitable for high-accuracy forecasting, whereas GRU is preferred for real-time or resource-constrained applications. Overall, this work highlights the potential of deep learning models to enhance the scalability, accuracy, and timeliness of intelligent river water quality monitoring systems.</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

Comparative study of deep learning models for multipoint biochemical oxygen demand prediction in urban rivers in Malaysia

  • 期刊论文

  • 发表日期：

    2026年2月18日

查看全文
  • N. M. Thamrin,
  • A. Jaffar,
  • M. S. A. Megat Ali,
  • I. M. Yassin,
  • M. F. Misnan,
  • K. G. Tiew

摘要

Managing river water quality by accurately predicting Biochemical Oxygen Demand (BOD) is essential for effective pollution control, particularly in urbanised catchments such as Malaysia's Klang River. Traditional BOD₅ methods are labour-intensive and unsuitable for real-time applications. This study aims to evaluate and compare the performance of three deep learning models, namely, Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Gated Recurrent Unit (GRU), for multipoint BOD prediction using eight physicochemical characteristics collected from eleven river monitoring stations. Data were processed using interpolation (PCHIP, Makima, and the Spline), and normalisation methods (Z-score and Min–Max) were used to form uniform time series for model training and validation. The results show that BiLSTM achieved the highest predictive accuracy with an average MAE of 0.3034 and RMSE of 0.3774, GRU recorded the fastest training time (18 s) while predictions (average MAE = 0.30 – 0.32 and RMSE = 0.37 – 0.39), GRU achieved the fastest training time (18 s) and the lowest training-phase errors (MAE = 0.0010, RMSE = 0.0018). In contrast, LSTM yielded slightly higher errors (MAE = 0.3217, RMSE = 0.4020). These findings demonstrate that BiLSTM is most suitable for high-accuracy forecasting, whereas GRU is preferred for real-time or resource-constrained applications. Overall, this work highlights the potential of deep learning models to enhance the scalability, accuracy, and timeliness of intelligent river water quality monitoring systems.