Scholarlink

<p>Accurate river flow simulation presents a fundamental challenge in water resource management, particularly given the complexities introduced by climate variability. While Long Short-Term Memory (LSTM) networks are powerful tools for modeling such temporal sequences, their performance is highly dependent on the precise tuning of hyperparameters and they are prone to overfitting. To address these limitations, this study proposes a novel hybrid framework that integrates the Block Bootstrap Aggregation (BBA) algorithm and population-based optimization algorithms (POAs) with an LSTM network. The objective of the proposed BBA-POA-LSTM model is to enhance the accuracy and stability of predictions under diverse climatic conditions. Within this framework, five metaheuristic algorithms—Harris Hawks Optimization (HHO), Whale Optimization Algorithm (WOA), Grasshopper Optimization Algorithm (GOA), Grey Wolf Optimization (GWO), and Particle Swarm Optimization (PSO)—are employed to optimize the critical hyperparameters of the LSTM, replacing traditional trial-and-error methods. Simultaneously, the BBA technique was utilized to generate synthetic datasets for model training, thereby mitigating overfitting and quantifying modeling uncertainty. The framework was rigorously evaluated using twenty years of hydro-climatic data (2004–2024) from five distinct hydrometric stations in Iran.The results demonstrate the superiority of the optimized hybrid models over the base trial-and-error LSTM model. Specifically, the WOA-LSTM model performed best for datasets DS1 and DS3, HHO-LSTM for DS2, GWO-LSTM for DS4, and GOA-LSTM for DS5. On average, these optimized models reduced the Root Mean Square Error (RMSE) by approximately 52 to 55% compared to the base LSTM model. Furthermore, the top-performing models achieved Nash–Sutcliffe Efficiency (NSE) index values exceeding 0.99 during the testing phase, indicating very high accuracy and an almost perfect fit between observed and simulated river flow values. A critical finding of this study is that no single optimization algorithm outperformed the others across all stations, underscoring the importance of selecting an optimizer tailored to the specific characteristics of the watershed. Ultimately, this research provides not only a precise mathematical model but also a practical decision-support tool for water resource managers, enabling them to conduct strategic planning with greater confidence in the face of climatic fluctuations.</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

Enhancing deep learning performance in hydrological forecasting: the role of population-based optimization and block bootstrapping

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月22日

查看全文
  • Adnan Mazraeh,
  • Meysam Bagherifar,
  • Saeid Shabanlou

摘要

Accurate river flow simulation presents a fundamental challenge in water resource management, particularly given the complexities introduced by climate variability. While Long Short-Term Memory (LSTM) networks are powerful tools for modeling such temporal sequences, their performance is highly dependent on the precise tuning of hyperparameters and they are prone to overfitting. To address these limitations, this study proposes a novel hybrid framework that integrates the Block Bootstrap Aggregation (BBA) algorithm and population-based optimization algorithms (POAs) with an LSTM network. The objective of the proposed BBA-POA-LSTM model is to enhance the accuracy and stability of predictions under diverse climatic conditions. Within this framework, five metaheuristic algorithms—Harris Hawks Optimization (HHO), Whale Optimization Algorithm (WOA), Grasshopper Optimization Algorithm (GOA), Grey Wolf Optimization (GWO), and Particle Swarm Optimization (PSO)—are employed to optimize the critical hyperparameters of the LSTM, replacing traditional trial-and-error methods. Simultaneously, the BBA technique was utilized to generate synthetic datasets for model training, thereby mitigating overfitting and quantifying modeling uncertainty. The framework was rigorously evaluated using twenty years of hydro-climatic data (2004–2024) from five distinct hydrometric stations in Iran.The results demonstrate the superiority of the optimized hybrid models over the base trial-and-error LSTM model. Specifically, the WOA-LSTM model performed best for datasets DS1 and DS3, HHO-LSTM for DS2, GWO-LSTM for DS4, and GOA-LSTM for DS5. On average, these optimized models reduced the Root Mean Square Error (RMSE) by approximately 52 to 55% compared to the base LSTM model. Furthermore, the top-performing models achieved Nash–Sutcliffe Efficiency (NSE) index values exceeding 0.99 during the testing phase, indicating very high accuracy and an almost perfect fit between observed and simulated river flow values. A critical finding of this study is that no single optimization algorithm outperformed the others across all stations, underscoring the importance of selecting an optimizer tailored to the specific characteristics of the watershed. Ultimately, this research provides not only a precise mathematical model but also a practical decision-support tool for water resource managers, enabling them to conduct strategic planning with greater confidence in the face of climatic fluctuations.