20200917_Hybrid regression model for near real-time urban water demand forecasting

1. Title, Journal and Authors

Title: Hybrid regression model for near real-time urban water demand forecasting

Journal: Computational and Applied Mathematics

Authors: Bruno M. Brentana, Edevar Luvizotto Jr.a , Manuel Herrerab, Joaquin Izquierdoc,*, Rafael Perez-Garciac


2. Summary


This paper was conducted to predict water consumption more accurately, and a hybrid model using Support Vector Regression (SVR) and adaptive Fourier series was presented.


As a result of comparing the water demand and time series data of each weather variable to determine the influence of the weather variable, the temperature and air humidity time series followed a regular distribution along with all time frames, but the wind speed and rain were irregular. In this comparison, it was possible to confirm a direct correlation between rain and air humidity, as well as temperature and water consumption. The collected data showed differences in water consumption and consumption patterns by month, day of the week, and time. These water consumption patterns can vary widely from region to region.


The predicted values ​​of the hybrid model using SVR and AFS were evaluated using the Root Mean Square Error (RMES), the mean absolute percentage error (MAE%), and the correlation coefficient (R^2). The RMSE of the SVR and SVR-AFS models were 1.318 and 4.767, respectively, and the MAE% were 3.45 and 12.91, respectively. Finally, R2 values ​​were 0.974 and 0.745. As a result, the hybrid model was able to predict water demand more accurately than the existing single model, but it was unable to capture the maximum and minimum values ​​well.


The proposed model could be an important tool for water utilities as the online function supports the operation and management of WDS and allows operators to program efficient operations to save energy and water.


3. Originality and Creativity


This study evaluated model efficiency about the size of the training data. Refreshing the model using 125 days of data showed the best efficiency.


4. Application to research



5. Contact

Yeong Gyu Gu / Intern student

Environmental Systems Engineering Lab.

School of Earth Sciences and Environmental EngineeringGwangju Institute of Science and Technology

1 Oryong-dong Buk-gu Gwangju, 500-712, Korea

Phone : +82-10-6589-6653

E-mail : kududrb1@naver.com

첨부 (0)