Work

Causality-Preserving Wavelet-Guided STL–VMD–Informer for Sustainable Urban Water-Supply Operations: Multi-Horizon Forecasting

Year: 2026

Type: article

Abstract: Reliable short-term water-supply forecasting supports sustainable urban water management by informing pump scheduling, water allocation, and operational planning. This study proposes a wavelet-guided STL–VMD–Informer framework for direct 1-, 3-, and 7-day forecasting while reducing look-ahead bias. Continuous wavelet transform analysis of the training subset identifies the dominant periodicity for STL, and VMD further decomposes the residual into band-limited modes. The training subset is initially decomposed, whereas validation and test features are updated sequentially using only observations available up to each forecast origin. The resulting components and calendar features are jointly modeled using a multivariate Informer. Using 581 daily SCADA observations from Degan Booster Pumping Station, the proposed model achieved an MAE of 4557m3d−1, an RMSE of 5891m3d−1, and an NSE of 0.803 for the 1-day horizon; NSE remained 0.631 and 0.524 for the 3- and 7-day horizons. External validation on public data from Liulin Water Treatment Plant yielded an MAE of 3540m3d−1 and an NSE of 0.857. The results demonstrate the potential of causality-aware multiscale forecasting for reliable and energy-conscious water-supply operation.

Source: Sustainability

DOI: 10.3390/su182010520

Authors: Hankang Lu, Yu Tian, Cheng Qian, Ruifang Yuan, Maoyuan Liu

Institutions: North China University of Water Resources and Electric Power, China Institute of Water Resources and Hydropower Research, Powerchina Huadong Engineering Corporation (China)

Language: English

Topic: Hydrological Forecasting Using AI

Subfield: Environmental Engineering

Field: Environmental Science

Domain: Physical Sciences

SDGs: Clean water and sanitation, Sustainable cities and communities

Open Access status: gold

Estimated APC paid: $2,896