Scholarlink

<p>This study proposes a novel multistage hybrid stochastic data-driven deep learning (DL) methodology with deep feature selection (DFS) for enhancing long-term seasonal streamflow prediction. The multistage hybrid MCMC-BC-DFS-BiLSTM-BiGRU model integrates bidirectional long short-term memory (BiLSTM) and bidirectional gated recurrent unit (BiGRU) architectures with Markov Chain Monte Carlo (MCMC)-based bivariate copulas (BC) within the DFS framework. Firstly, a DFS strategy was developed using the Ant Colony Optimization (ACO) algorithm, autocorrelation function (ACF), partial autocorrelation function (PACF), and recursive feature elimination (RFE) method to determine optimized predictor variables (PV) from a pool of three predictor candidates, i.e., local meteorological variables, large-scale atmospheric indices, and solar activity indices. Subsequently, several diverse MCMC-BC models were employed using optimal PV obtained through the DFS framework (MCMC-BC-DFS) to evaluate the connection between the current seasonal streamflow and its potential future variations. Finally, the most suitable MCMC-BC-DFS model was incorporated into the hybrid BiLSTM-BiGRU model to predict spring (Sep–Nov) streamflow across nine catchments in the Victorian site of the Upper Murray Basin (UMB), Australia. The proposed multistage hybrid MCMC-BC-DFS-BiLSTM-BiGRU model was compared to several machine learning (ML) models (multilayer perceptron (MLP), extreme gradient boosting (XGBoost), support vector machine (SVM), and random forest (RF)) through different robust statistical metrics and graphical illustrations. According to experimental findings, the proposed model demonstrated excellent performance in long-term streamflow prediction at the seasonal timescale by outperforming its benchmark counterparts based on all statistical indicators.&#xa0;Consequently, as a pioneer study, the proposed multistage hybrid model can be effectively utilized as a highly promising tool for upstream seasonal predictions that could be helpful to policymakers for more informed environmental decision-making.</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 seasonal streamflow prediction using multistage hybrid stochastic data-driven deep learning methodology with deep feature selection

  • 期刊论文

  • 发表日期：

    2025年2月23日

查看全文
  • Asif Iqbal,
  • Tanveer Ahmed Siddiqi

摘要

This study proposes a novel multistage hybrid stochastic data-driven deep learning (DL) methodology with deep feature selection (DFS) for enhancing long-term seasonal streamflow prediction. The multistage hybrid MCMC-BC-DFS-BiLSTM-BiGRU model integrates bidirectional long short-term memory (BiLSTM) and bidirectional gated recurrent unit (BiGRU) architectures with Markov Chain Monte Carlo (MCMC)-based bivariate copulas (BC) within the DFS framework. Firstly, a DFS strategy was developed using the Ant Colony Optimization (ACO) algorithm, autocorrelation function (ACF), partial autocorrelation function (PACF), and recursive feature elimination (RFE) method to determine optimized predictor variables (PV) from a pool of three predictor candidates, i.e., local meteorological variables, large-scale atmospheric indices, and solar activity indices. Subsequently, several diverse MCMC-BC models were employed using optimal PV obtained through the DFS framework (MCMC-BC-DFS) to evaluate the connection between the current seasonal streamflow and its potential future variations. Finally, the most suitable MCMC-BC-DFS model was incorporated into the hybrid BiLSTM-BiGRU model to predict spring (Sep–Nov) streamflow across nine catchments in the Victorian site of the Upper Murray Basin (UMB), Australia. The proposed multistage hybrid MCMC-BC-DFS-BiLSTM-BiGRU model was compared to several machine learning (ML) models (multilayer perceptron (MLP), extreme gradient boosting (XGBoost), support vector machine (SVM), and random forest (RF)) through different robust statistical metrics and graphical illustrations. According to experimental findings, the proposed model demonstrated excellent performance in long-term streamflow prediction at the seasonal timescale by outperforming its benchmark counterparts based on all statistical indicators. Consequently, as a pioneer study, the proposed multistage hybrid model can be effectively utilized as a highly promising tool for upstream seasonal predictions that could be helpful to policymakers for more informed environmental decision-making.