Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Comparison of ARIMA and NNAR Models for Forecasting Water Treatment Plant’s Influent Characteristics
Abstract A reliable forecasting model for each Water Treatment Plant (WTP) influent characteristics is useful for controlling the plant’s operation. In this paper Auto-Regressive Integrated Moving Average (ARIMA) and Neural Network Auto-Regressive (NNAR) modeling techniques were applied on a WTP’s influent water characteristics time series to make some models for short-term period (to seven days ahead) forecasting. The ARIMA and NNAR models both provided acceptable generalization capability with R2s ranged from 0.44 to 0.91 and 0.45 to 0.92, respectively, for chloride and temperature. Although a more prediction performance was observed for NNAR in comparison with ARIMA for all studied series, the forecasting performance of models was further examined using Time Series Cross-Validation (TSCV) and Diebold-Mariano test. The results showed ARIMA is more accurate than NNAR for forecasting the horizon-daily values for CO2, Cl and Ca time-series. Therefore, despite of the good predictive performance of NNAR, ARIMA may still stands as better alternative for forecasting task of aforementioned series. Thus, as a general rule, not only the predictive performance using R2 statistic but also the forecasting performance of a model using TSCV, are need to be examined and compared for selecting an appropriate forecasting model for WTP’s influent characteristics.
Comparison of ARIMA and NNAR Models for Forecasting Water Treatment Plant’s Influent Characteristics
Abstract A reliable forecasting model for each Water Treatment Plant (WTP) influent characteristics is useful for controlling the plant’s operation. In this paper Auto-Regressive Integrated Moving Average (ARIMA) and Neural Network Auto-Regressive (NNAR) modeling techniques were applied on a WTP’s influent water characteristics time series to make some models for short-term period (to seven days ahead) forecasting. The ARIMA and NNAR models both provided acceptable generalization capability with R2s ranged from 0.44 to 0.91 and 0.45 to 0.92, respectively, for chloride and temperature. Although a more prediction performance was observed for NNAR in comparison with ARIMA for all studied series, the forecasting performance of models was further examined using Time Series Cross-Validation (TSCV) and Diebold-Mariano test. The results showed ARIMA is more accurate than NNAR for forecasting the horizon-daily values for CO2, Cl and Ca time-series. Therefore, despite of the good predictive performance of NNAR, ARIMA may still stands as better alternative for forecasting task of aforementioned series. Thus, as a general rule, not only the predictive performance using R2 statistic but also the forecasting performance of a model using TSCV, are need to be examined and compared for selecting an appropriate forecasting model for WTP’s influent characteristics.
Comparison of ARIMA and NNAR Models for Forecasting Water Treatment Plant’s Influent Characteristics
Maleki, Afshin (Autor:in) / Nasseri, Simin (Autor:in) / Aminabad, Mehri Solaimany (Autor:in) / Hadi, Mahdi (Autor:in)
KSCE Journal of Civil Engineering ; 22 ; 3233-3245
10.04.2018
13 pages
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
Comparison of ARIMA and NNAR Models for Forecasting Water Treatment Plant’s Influent Characteristics
| Springer Verlag | 2018
Comparison of ARIMA and NNAR Models for Forecasting Water Treatment Plant’s Influent Characteristics
| Online Contents | 2018
Istanbul Dam Water Levels Forecasting Using ARIMA Models
| IEEE | 2022
| DOAJ | 2018