Scholarlink

<p>Predicting river flows is crucial for watershed management in different regions of the world. Therefore, our study proposes the CEEMDAN- bidirectional long short-term memory (BILSTM)- artificial neural network (ANN) and CEEMDAN-BILSTM-SVM models to predict one-day-ahead river flow in Ajichay River, Iran. The new model uses CEEMDAN and BILSTM to reduce the complexity of the time series and capture time series features. The models also use meteorological parameters and lagged river flow data as input data. The data set was collected from 2014 to 2019. The study uses different performance metrics to compare the new model with other predictive models. The CEEMDAN-BILSTM-ANN model has performance metrics of NSE = 0.97, Kling–Gupta Efficiency = 0.95, relative root mean square percent error = 12, mean absolute error = 0.125, and standard deviation of relative error = 2.12. Our study also uses generalized likelihood uncertainty estimation (GLUE) to quantify the uncertainty of model outputs. The study results indicate that the outputs of the CEEMDAN-BILSTM-ANN and CEEMDAN-BILSTM-SVM models have lower uncertainty than the other models. Our study also sets the model parameters using the mother optimization algorithm. The algorithm avoids the problem of getting trapped in the local optima and effectively adjusts the model parameters (e.g., CEEMDAN, SVM, and BILSTM parameters). Our results also show that CEEMDAN improves the computational efficiency of models by producing time series with simpler patterns.</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

CEEMDAN-BILSTM-ANN and SVM Models: Two Robust Predictive Models for Predicting River flow

  • 期刊论文

  • 发表日期：

    2025年1月22日

查看全文
  • Elham Ghanbari-Adivi,
  • Mohammad Ehteram

摘要

Predicting river flows is crucial for watershed management in different regions of the world. Therefore, our study proposes the CEEMDAN- bidirectional long short-term memory (BILSTM)- artificial neural network (ANN) and CEEMDAN-BILSTM-SVM models to predict one-day-ahead river flow in Ajichay River, Iran. The new model uses CEEMDAN and BILSTM to reduce the complexity of the time series and capture time series features. The models also use meteorological parameters and lagged river flow data as input data. The data set was collected from 2014 to 2019. The study uses different performance metrics to compare the new model with other predictive models. The CEEMDAN-BILSTM-ANN model has performance metrics of NSE = 0.97, Kling–Gupta Efficiency = 0.95, relative root mean square percent error = 12, mean absolute error = 0.125, and standard deviation of relative error = 2.12. Our study also uses generalized likelihood uncertainty estimation (GLUE) to quantify the uncertainty of model outputs. The study results indicate that the outputs of the CEEMDAN-BILSTM-ANN and CEEMDAN-BILSTM-SVM models have lower uncertainty than the other models. Our study also sets the model parameters using the mother optimization algorithm. The algorithm avoids the problem of getting trapped in the local optima and effectively adjusts the model parameters (e.g., CEEMDAN, SVM, and BILSTM parameters). Our results also show that CEEMDAN improves the computational efficiency of models by producing time series with simpler patterns.