Scholarlink

<p>Accurate prediction of daily reservoir inflows is essential for effective water resource management, particularly in regions with variable water availability. This study evaluates the performance of standalone (GRU, LSTM, TCN) and hybrid machine learning models (TVF-EMD-GRU, TVF-EMD-LSTM, TVF-EMD-TCN) in forecasting inflows at the Dez and MasjedSoleyman reservoirs. The hybrid models incorporate Time-Varying Filter-based Empirical Mode Decomposition (TVF-EMD) to handle data complexity, with sample entropy applied to assess the regularity and predictability of the decomposed components. Additionally, the Holt-Winters method is applied to stationary components for capturing seasonal variations. The TVF-EMD-TCN model consistently outperforms other models, achieving 93% of predictions within the 0–5&#xa0;m³/s error range for the Dez reservoir and 92% for MasjedSoleyman, effectively minimizing larger errors. In comparison, standalone models, particularly GRU and LSTM, display higher frequencies of errors in the 5–10&#xa0;m³/s range, especially for MasjedSoleyman. The hybrid models, especially TVF-EMD-TCN, demonstrate robust performance across different hydrological conditions, with TVF-EMD-GRU and TVF-EMD-LSTM also showing improved accuracy, capturing 88–91% of errors within the smallest range for both reservoirs. These findings underscore the superiority of the hybrid models, particularly TVF-EMD-TCN, in providing reliable predicts, crucial for the efficient management of water resources.</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

Multi-step Daily Reservoir Inflow Prediction using Temporal Conventional Network Model Integrating time-Varying Filter-Based empirical mode Decomposition and Holt-Winters Algorithms

  • 期刊论文

  • 发表日期：

    2026年6月27日

查看全文
  • Atefeh Sadat Hosseini,
  • Sultan K. Salamah,
  • Ahmed Alkhayyat,
  • Anupam Yadav,
  • Abhinav Kumar,
  • Marwan Kheimi,
  • Sandeep Singh,
  • KDV Prasad

摘要

Accurate prediction of daily reservoir inflows is essential for effective water resource management, particularly in regions with variable water availability. This study evaluates the performance of standalone (GRU, LSTM, TCN) and hybrid machine learning models (TVF-EMD-GRU, TVF-EMD-LSTM, TVF-EMD-TCN) in forecasting inflows at the Dez and MasjedSoleyman reservoirs. The hybrid models incorporate Time-Varying Filter-based Empirical Mode Decomposition (TVF-EMD) to handle data complexity, with sample entropy applied to assess the regularity and predictability of the decomposed components. Additionally, the Holt-Winters method is applied to stationary components for capturing seasonal variations. The TVF-EMD-TCN model consistently outperforms other models, achieving 93% of predictions within the 0–5 m³/s error range for the Dez reservoir and 92% for MasjedSoleyman, effectively minimizing larger errors. In comparison, standalone models, particularly GRU and LSTM, display higher frequencies of errors in the 5–10 m³/s range, especially for MasjedSoleyman. The hybrid models, especially TVF-EMD-TCN, demonstrate robust performance across different hydrological conditions, with TVF-EMD-GRU and TVF-EMD-LSTM also showing improved accuracy, capturing 88–91% of errors within the smallest range for both reservoirs. These findings underscore the superiority of the hybrid models, particularly TVF-EMD-TCN, in providing reliable predicts, crucial for the efficient management of water resources.