Investigating capabilities of machine learning techniques in forecasting streamflow

Research output: Contribution to journalArticle

Abstract

This paper presents a systematic investigation into modelling capacities of three conventional data driven modelling techniques, namely, wavelet based artificial neural network (WANN), support vector regression (SVR), and deep belief network (DBN) for multi-step ahead stream flow forecasting. To evaluate the effectiveness of these modelling techniques, hydro-meteorological hourly datasets from three case-study rivers located in the UK have been used. A heuristic performance analysis of the modelling schemes has been conducted by systematically analysing the key statistics that measure magnitude, scatter and density of model errors. Finally, for each of the modelling technique, performance deterioration rate in time was estimated. The results show that the SVR model can forecast quite accurately up to one to two hours ahead but its performance deteriorates gradually from three hour onwards. Further it has been found that the WANN model performs better when the overall nonlinearity of the system increases whereas the DBN model appeared to show consistent poor predictive capabilities when compared to the other models presented herein. We conclude by stating that for any selected model, it is possible to use identical model structure for up to two step ahead forecasting. Models need to be re-configured beyond that limit.
Original languageEnglish
JournalProceedings of the ICE - Water Management
Early online date28 May 2019
DOIs
Publication statusE-pub ahead of print - 28 May 2019

Fingerprint

Learning systems
Bayesian networks
Neural networks
Stream flow
Model structures
Deterioration
Data structures
Rivers
Statistics

Keywords

  • Hydrology and water resource
  • Computational mechanics
  • Statistical analysis

Cite this

@article{7556c7cdd45f48aeaf5e8f1e462780b2,
title = "Investigating capabilities of machine learning techniques in forecasting streamflow",
abstract = "This paper presents a systematic investigation into modelling capacities of three conventional data driven modelling techniques, namely, wavelet based artificial neural network (WANN), support vector regression (SVR), and deep belief network (DBN) for multi-step ahead stream flow forecasting. To evaluate the effectiveness of these modelling techniques, hydro-meteorological hourly datasets from three case-study rivers located in the UK have been used. A heuristic performance analysis of the modelling schemes has been conducted by systematically analysing the key statistics that measure magnitude, scatter and density of model errors. Finally, for each of the modelling technique, performance deterioration rate in time was estimated. The results show that the SVR model can forecast quite accurately up to one to two hours ahead but its performance deteriorates gradually from three hour onwards. Further it has been found that the WANN model performs better when the overall nonlinearity of the system increases whereas the DBN model appeared to show consistent poor predictive capabilities when compared to the other models presented herein. We conclude by stating that for any selected model, it is possible to use identical model structure for up to two step ahead forecasting. Models need to be re-configured beyond that limit.",
keywords = "Hydrology and water resource, Computational mechanics, Statistical analysis",
author = "Syed Kabir and Sandhya Patidar and Gareth Pender",
year = "2019",
month = "5",
day = "28",
doi = "10.1680/jwama.19.00001",
language = "English",
journal = "Proceedings of the ICE - Water Management",
issn = "1741-7589",
publisher = "ICE Publishing Ltd.",

}

TY - JOUR

T1 - Investigating capabilities of machine learning techniques in forecasting streamflow

AU - Kabir, Syed

AU - Patidar, Sandhya

AU - Pender, Gareth

PY - 2019/5/28

Y1 - 2019/5/28

N2 - This paper presents a systematic investigation into modelling capacities of three conventional data driven modelling techniques, namely, wavelet based artificial neural network (WANN), support vector regression (SVR), and deep belief network (DBN) for multi-step ahead stream flow forecasting. To evaluate the effectiveness of these modelling techniques, hydro-meteorological hourly datasets from three case-study rivers located in the UK have been used. A heuristic performance analysis of the modelling schemes has been conducted by systematically analysing the key statistics that measure magnitude, scatter and density of model errors. Finally, for each of the modelling technique, performance deterioration rate in time was estimated. The results show that the SVR model can forecast quite accurately up to one to two hours ahead but its performance deteriorates gradually from three hour onwards. Further it has been found that the WANN model performs better when the overall nonlinearity of the system increases whereas the DBN model appeared to show consistent poor predictive capabilities when compared to the other models presented herein. We conclude by stating that for any selected model, it is possible to use identical model structure for up to two step ahead forecasting. Models need to be re-configured beyond that limit.

AB - This paper presents a systematic investigation into modelling capacities of three conventional data driven modelling techniques, namely, wavelet based artificial neural network (WANN), support vector regression (SVR), and deep belief network (DBN) for multi-step ahead stream flow forecasting. To evaluate the effectiveness of these modelling techniques, hydro-meteorological hourly datasets from three case-study rivers located in the UK have been used. A heuristic performance analysis of the modelling schemes has been conducted by systematically analysing the key statistics that measure magnitude, scatter and density of model errors. Finally, for each of the modelling technique, performance deterioration rate in time was estimated. The results show that the SVR model can forecast quite accurately up to one to two hours ahead but its performance deteriorates gradually from three hour onwards. Further it has been found that the WANN model performs better when the overall nonlinearity of the system increases whereas the DBN model appeared to show consistent poor predictive capabilities when compared to the other models presented herein. We conclude by stating that for any selected model, it is possible to use identical model structure for up to two step ahead forecasting. Models need to be re-configured beyond that limit.

KW - Hydrology and water resource

KW - Computational mechanics

KW - Statistical analysis

U2 - 10.1680/jwama.19.00001

DO - 10.1680/jwama.19.00001

M3 - Article

JO - Proceedings of the ICE - Water Management

JF - Proceedings of the ICE - Water Management

SN - 1741-7589

ER -