| Title |
Developing a Return Flow Prediction Model for Residential Water Use under a Water-Cycle System Perspective: Case Study of Jangseong and Damyang |
| Authors |
이승연(Lee, Seoungyeon);이상은(Lee, Sangeun) |
| DOI |
https://doi.org/10.12652/Ksce.2025.45.6.0765 |
| Keywords |
물순환시스템; 통계 모형; 기계학습 모형; 재귀예측 Water-cycle system; Statistical model; Machine-learning model; Recursive forecasting |
| Abstract |
This study developed a basin-centered, areal analysis technique for residential water use to support sustainable water management. In contrast to conventional discharge prediction studies that treat each monitoring point independently, this research advances a unified modeling framework that integrates the full domestic water-cycle system and its underlying interdependencies. As a follow-up to Lee and Lee(2023), we constructed a water-cycle system for the Jangseong?Damyang(Jeollanam-do) with a more complex water-supply and sewer network. Focusing on complete measurement points, the system was defined as three inflow sites, four outflow sites, and the analysis period was 2020-01-01?2024-05-31(daily). The dataset was split into 70 % training and 30 % validation, and applied
statistical models(TFM, DRM) and machine-learning models(GBR, Random Forest, Ridge regression model). Ridge regression achieved superior performance at most sites. To analyze the applicability across different lead times, recursive forecasting was applied. As a result, Y1 and Y4 were usable for Lead-1 to Lead-7 with TFM and Ridge; Y2 for Lead-1 with the Ridge; and Y3 with the Ridge for short leads and TFM for longer leads. This study integrates input?output lag structures to demonstrate improved potential for flow monitoring and prediction. Future work should incorporate exogenous variables, include incomplete measurement points, and extend
to weekly and monthly time scales to better support mid- to long-term operational decision-making. |