24-Hours Demand Forecasting Based on SARIMA and Support Vector Machines

Applying human mobility and water consumption data for short-term water demand forecasting using classical and machine learning models

“…where η are the coefficients of external inputs estimated at a model fitting step. ARIMA and ARIMAX models have been used for decades and valued for their accuracy (Braun et al 2014;Alvisi, Franchini, and Marinelli 2007). Various extensions of the models were proposed, one of which are SARIMA (seasonal ARIMA) models that account for seasonal effects approximated from the past data, improving their accuracy in long-term forecasting.…”

“…Also, Ghalehkhondabi et al (2017) in the review paper point out that the machine learning (named soft computing methods) are outperforming classical multilinear regression, multiple non-linear regression and ARIMA methods. Literature review (Braun et al 2014;Chen et al 2017;Antunes et al 2018) offers usually a comparison of shorter forecasts: daily water demand/use (one value per day) or sub-daily prediction (24 values per day). In these studies SVR (Braun et al 2014) is outperforming SARIMA by 1% (1-sigma) and 4% (2-sigma) (measured as a percentile relative error).…”

“…The most commonly used predictive models for water demand forecasting comprise autoregressive integrated moving average (ARIMA) and autoregressive integrated moving average with exogenous variables (ARIMAX) methods (Braun et al 2014;Alvisi, Franchini, and Marinelli 2007). Recent studies, however, introduce machine learning (ML) solutions, such as Artificial Neural Networks (ANN), Support Vector Regression (SVR) and regression trees, which now outperform the classical methods (Herrera et al 2010;Tiwari and Adamowski 2017;Bennett, Stewart, and Beal 2013;Bai et al 2014;Donkor et al 2014;Antunes et al 2018;Xu et al 2019).…”

A Spectral Approach to Uncertainty Quantification in Water Distribution Networks

“…For fast changing parameters like the nodal demand there exist models that try to predict the value for the next time step. This may be achieved either by data driven models Herrera et al (2010); Braun et al (2014) or physical models that simulate the stochastic nature of consumers Blokker and Van der Schee (2006). Other applications choose to solve the problem through calibration methods, which try to identify a set of parameters that give the best fit between measured and simulated data.…”

The Analysis and Forecasting of Time Sequence Based on SARIMA Model

“…And the water demand in Iran was estimated based on SARIMA model [3]. In theory, SARIMA model is suitable for the forecasting of time sequences with nonstationarity and seasonality [4][5]. In this paper, we establish a SARIMA model to forecast the quantities of commodity barcodes registration in the future.…”