Harmonize input selection for sediment transport prediction

Haitham Abdulmohsin Afan, Behrooz Keshtegar, Wan Hanna Melini Wan Mohtar, Ahmed El-Shafie

Research output: Contribution to journalArticle

9 Citations (Scopus)

Abstract

In this paper, three modeling approaches using a Neural Network (NN), Response Surface Method (RSM) and response surface method basis Global Harmony Search (GHS) are applied to predict the daily time series suspended sediment load. Generally, the input variables for forecasting the suspended sediment load are manually selected based on the maximum correlations of input variables in the modeling approaches based on NN and RSM. The RSM is improved to select the input variables by using the errors terms of training data based on the GHS, namely as response surface method and global harmony search (RSM-GHS) modeling method. The second-order polynomial function with cross terms is applied to calibrate the time series suspended sediment load with three, four and five input variables in the proposed RSM-GHS. The linear, square and cross corrections of twenty input variables of antecedent values of suspended sediment load and water discharge are investigated to achieve the best predictions of the RSM based on the GHS method. The performances of the NN, RSM and proposed RSM-GHS including both accuracy and simplicity are compared through several comparative predicted and error statistics. The results illustrated that the proposed RSM-GHS is as uncomplicated as the RSM but performed better, where fewer errors and better correlation was observed (R = 0.95, MAE = 18.09 (ton/day), RMSE = 25.16 (ton/day)) compared to the ANN (R = 0.91, MAE = 20.17 (ton/day), RMSE = 33.09 (ton/day)) and RSM (R = 0.91, MAE = 20.06 (ton/day), RMSE = 31.92 (ton/day)) for all types of input variables.

Original languageEnglish
Pages (from-to)366-375
Number of pages10
JournalJournal of Hydrology
Volume552
DOIs
Publication statusPublished - 1 Sep 2017

Fingerprint

sediment transport
prediction
suspended sediment
method
time series
modeling

Keywords

  • Global-harmony search
  • Neural network
  • Response surface method
  • Suspended sediment load
  • Tropical river

ASJC Scopus subject areas

  • Water Science and Technology

Cite this

Harmonize input selection for sediment transport prediction. / Afan, Haitham Abdulmohsin; Keshtegar, Behrooz; Wan Mohtar, Wan Hanna Melini; El-Shafie, Ahmed.

In: Journal of Hydrology, Vol. 552, 01.09.2017, p. 366-375.

Research output: Contribution to journalArticle

Afan, Haitham Abdulmohsin ; Keshtegar, Behrooz ; Wan Mohtar, Wan Hanna Melini ; El-Shafie, Ahmed. / Harmonize input selection for sediment transport prediction. In: Journal of Hydrology. 2017 ; Vol. 552. pp. 366-375.
@article{93be48dc959a4c81bbde1d21dd74e33c,
title = "Harmonize input selection for sediment transport prediction",
abstract = "In this paper, three modeling approaches using a Neural Network (NN), Response Surface Method (RSM) and response surface method basis Global Harmony Search (GHS) are applied to predict the daily time series suspended sediment load. Generally, the input variables for forecasting the suspended sediment load are manually selected based on the maximum correlations of input variables in the modeling approaches based on NN and RSM. The RSM is improved to select the input variables by using the errors terms of training data based on the GHS, namely as response surface method and global harmony search (RSM-GHS) modeling method. The second-order polynomial function with cross terms is applied to calibrate the time series suspended sediment load with three, four and five input variables in the proposed RSM-GHS. The linear, square and cross corrections of twenty input variables of antecedent values of suspended sediment load and water discharge are investigated to achieve the best predictions of the RSM based on the GHS method. The performances of the NN, RSM and proposed RSM-GHS including both accuracy and simplicity are compared through several comparative predicted and error statistics. The results illustrated that the proposed RSM-GHS is as uncomplicated as the RSM but performed better, where fewer errors and better correlation was observed (R = 0.95, MAE = 18.09 (ton/day), RMSE = 25.16 (ton/day)) compared to the ANN (R = 0.91, MAE = 20.17 (ton/day), RMSE = 33.09 (ton/day)) and RSM (R = 0.91, MAE = 20.06 (ton/day), RMSE = 31.92 (ton/day)) for all types of input variables.",
keywords = "Global-harmony search, Neural network, Response surface method, Suspended sediment load, Tropical river",
author = "Afan, {Haitham Abdulmohsin} and Behrooz Keshtegar and {Wan Mohtar}, {Wan Hanna Melini} and Ahmed El-Shafie",
year = "2017",
month = "9",
day = "1",
doi = "10.1016/j.jhydrol.2017.07.008",
language = "English",
volume = "552",
pages = "366--375",
journal = "Journal of Hydrology",
issn = "0022-1694",
publisher = "Elsevier",

}

TY - JOUR

T1 - Harmonize input selection for sediment transport prediction

AU - Afan, Haitham Abdulmohsin

AU - Keshtegar, Behrooz

AU - Wan Mohtar, Wan Hanna Melini

AU - El-Shafie, Ahmed

PY - 2017/9/1

Y1 - 2017/9/1

N2 - In this paper, three modeling approaches using a Neural Network (NN), Response Surface Method (RSM) and response surface method basis Global Harmony Search (GHS) are applied to predict the daily time series suspended sediment load. Generally, the input variables for forecasting the suspended sediment load are manually selected based on the maximum correlations of input variables in the modeling approaches based on NN and RSM. The RSM is improved to select the input variables by using the errors terms of training data based on the GHS, namely as response surface method and global harmony search (RSM-GHS) modeling method. The second-order polynomial function with cross terms is applied to calibrate the time series suspended sediment load with three, four and five input variables in the proposed RSM-GHS. The linear, square and cross corrections of twenty input variables of antecedent values of suspended sediment load and water discharge are investigated to achieve the best predictions of the RSM based on the GHS method. The performances of the NN, RSM and proposed RSM-GHS including both accuracy and simplicity are compared through several comparative predicted and error statistics. The results illustrated that the proposed RSM-GHS is as uncomplicated as the RSM but performed better, where fewer errors and better correlation was observed (R = 0.95, MAE = 18.09 (ton/day), RMSE = 25.16 (ton/day)) compared to the ANN (R = 0.91, MAE = 20.17 (ton/day), RMSE = 33.09 (ton/day)) and RSM (R = 0.91, MAE = 20.06 (ton/day), RMSE = 31.92 (ton/day)) for all types of input variables.

AB - In this paper, three modeling approaches using a Neural Network (NN), Response Surface Method (RSM) and response surface method basis Global Harmony Search (GHS) are applied to predict the daily time series suspended sediment load. Generally, the input variables for forecasting the suspended sediment load are manually selected based on the maximum correlations of input variables in the modeling approaches based on NN and RSM. The RSM is improved to select the input variables by using the errors terms of training data based on the GHS, namely as response surface method and global harmony search (RSM-GHS) modeling method. The second-order polynomial function with cross terms is applied to calibrate the time series suspended sediment load with three, four and five input variables in the proposed RSM-GHS. The linear, square and cross corrections of twenty input variables of antecedent values of suspended sediment load and water discharge are investigated to achieve the best predictions of the RSM based on the GHS method. The performances of the NN, RSM and proposed RSM-GHS including both accuracy and simplicity are compared through several comparative predicted and error statistics. The results illustrated that the proposed RSM-GHS is as uncomplicated as the RSM but performed better, where fewer errors and better correlation was observed (R = 0.95, MAE = 18.09 (ton/day), RMSE = 25.16 (ton/day)) compared to the ANN (R = 0.91, MAE = 20.17 (ton/day), RMSE = 33.09 (ton/day)) and RSM (R = 0.91, MAE = 20.06 (ton/day), RMSE = 31.92 (ton/day)) for all types of input variables.

KW - Global-harmony search

KW - Neural network

KW - Response surface method

KW - Suspended sediment load

KW - Tropical river

UR - http://www.scopus.com/inward/record.url?scp=85023644667&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85023644667&partnerID=8YFLogxK

U2 - 10.1016/j.jhydrol.2017.07.008

DO - 10.1016/j.jhydrol.2017.07.008

M3 - Article

VL - 552

SP - 366

EP - 375

JO - Journal of Hydrology

JF - Journal of Hydrology

SN - 0022-1694

ER -