TY - JOUR
T1 - An enhanced extreme learning machine model for river flow forecasting
T2 - State-of-the-art, practical applications in water resource engineering area and future research direction
AU - Yaseen, Zaher Mundher
AU - Sulaiman, Sadeq Oleiwi
AU - Deo, Ravinesh C.
AU - Chau, Kwok Wing
PY - 2019/2
Y1 - 2019/2
N2 - Despite the massive diversity in the modeling requirements for practical hydrological applications, there remains a need to develop more reliable and intelligent expert systems used for real-time prediction purposes. The challenge in meeting the standards of an expert system is primarily due to the influence and behavior of hydrological processes that is driven by natural fluctuations over the physical scale, and the resulting variance in the underlying model input datasets. River flow forecasting is an imperative task for water resources operation and management, water demand assessments, irrigation and agriculture, early flood warning and hydropower generations. This paper aims to investigate the viability of the enhanced version of extreme learning machine (EELM) model in river flow forecasting applied in a tropical environment. Herein, we apply the complete orthogonal decomposition (COD) learning tool to tune the output-hidden layer of the ELM model's internal neuronal system, instead of the conventional multi-resolution tool (e.g., singular value decomposition). To demonstrate the application of EELM model, the Kelantan River, located in the Malaysian peninsular, selected as a case study. For a comparison of the EELM model, and further model evaluation, two distinct data-intelligent models are developed (i.e., the classical ELM and the support vector regression, SVR model). An exhaustive list of diagnostic indicators are used to evaluate the EELM model in respect to the benchmark algorithms, namely, SVR and ELM. The model performance indicators exhibit superior results for the EELM model relative to ELM and SVR models. In addition, the EELM model is presented as a more accurate, alternative predictive tool for modelling the tropical river flow patterns and its underlying characteristic perturbations in the physical space. Several statistical metrics defined as the coefficient of determination (r), Nash-Sutcliffe efficiency (Ens), Willmott's Index (WI), root-mean-square error (RMSE) and mean absolute error (MAE) are computed to assess the model's effectiveness. In quantitative terms, superiority of EELM over ELM and SVR models was exhibited by Ens = 0.7995, 0.7434 and 0.665, r = 0.894, 0.869 and 0.818 and WI = 0.9380, 0.9180 and 0.8921, respectively. Whereas, EELM model attained lower (RMSE and MAE) values by approximately (11.61–22.53%) and (8.26–8.72%) relative to ELM and SVR models, respectively. The obtained results reveal that the EELM model is a robust expert model and can be embraced practically in real-life water resources management and river sustainability decisions. As a complementary component of this paper, we also review state-of-art research works where scholars have embraced extensive implementation of the ELM model in water resource engineering problems. A comprehensive evaluation is carried out to recognize the current limitations, and also to propose potential opportunities of applying improved variants of the ELM model presented as a future research direction.
AB - Despite the massive diversity in the modeling requirements for practical hydrological applications, there remains a need to develop more reliable and intelligent expert systems used for real-time prediction purposes. The challenge in meeting the standards of an expert system is primarily due to the influence and behavior of hydrological processes that is driven by natural fluctuations over the physical scale, and the resulting variance in the underlying model input datasets. River flow forecasting is an imperative task for water resources operation and management, water demand assessments, irrigation and agriculture, early flood warning and hydropower generations. This paper aims to investigate the viability of the enhanced version of extreme learning machine (EELM) model in river flow forecasting applied in a tropical environment. Herein, we apply the complete orthogonal decomposition (COD) learning tool to tune the output-hidden layer of the ELM model's internal neuronal system, instead of the conventional multi-resolution tool (e.g., singular value decomposition). To demonstrate the application of EELM model, the Kelantan River, located in the Malaysian peninsular, selected as a case study. For a comparison of the EELM model, and further model evaluation, two distinct data-intelligent models are developed (i.e., the classical ELM and the support vector regression, SVR model). An exhaustive list of diagnostic indicators are used to evaluate the EELM model in respect to the benchmark algorithms, namely, SVR and ELM. The model performance indicators exhibit superior results for the EELM model relative to ELM and SVR models. In addition, the EELM model is presented as a more accurate, alternative predictive tool for modelling the tropical river flow patterns and its underlying characteristic perturbations in the physical space. Several statistical metrics defined as the coefficient of determination (r), Nash-Sutcliffe efficiency (Ens), Willmott's Index (WI), root-mean-square error (RMSE) and mean absolute error (MAE) are computed to assess the model's effectiveness. In quantitative terms, superiority of EELM over ELM and SVR models was exhibited by Ens = 0.7995, 0.7434 and 0.665, r = 0.894, 0.869 and 0.818 and WI = 0.9380, 0.9180 and 0.8921, respectively. Whereas, EELM model attained lower (RMSE and MAE) values by approximately (11.61–22.53%) and (8.26–8.72%) relative to ELM and SVR models, respectively. The obtained results reveal that the EELM model is a robust expert model and can be embraced practically in real-life water resources management and river sustainability decisions. As a complementary component of this paper, we also review state-of-art research works where scholars have embraced extensive implementation of the ELM model in water resource engineering problems. A comprehensive evaluation is carried out to recognize the current limitations, and also to propose potential opportunities of applying improved variants of the ELM model presented as a future research direction.
KW - Complete orthogonal decomposition
KW - Extreme learning machine
KW - Future research direction
KW - River flow forecasting
KW - State-of-the-art
KW - Water resources engineering
UR - http://www.scopus.com/inward/record.url?scp=85059162436&partnerID=8YFLogxK
U2 - 10.1016/j.jhydrol.2018.11.069
DO - 10.1016/j.jhydrol.2018.11.069
M3 - Review article
AN - SCOPUS:85059162436
SN - 0022-1694
VL - 569
SP - 387
EP - 408
JO - Journal of Hydrology
JF - Journal of Hydrology
ER -