Development of water quality prediction model using LTSF-Linear and complete ensemble empirical mode decomposition

Abstract

South Korea's water supply system has a high coverage rate, still, it faces significant challenges in effective water quality (WQ) management due to seasonal variations in source water turbidity, frequent algal blooms, and aging water treatment facilities and pipelines. In particular, climate change and shifting rainfall patterns are increasing the variability of raw water quality, which is a major factor challenging the water treatment process. These issues not only reduce water treatment efficiency but also increase the risk of WQ contamination incidents, which can adversely affect public health. Therefore, rapid and accurate prediction of key WQ indicators such as pH, EC is essential for proactive WQ management and WQ contamination response. This study proposes a more effective WQ management approach by utilizing real-time WQ. To achieve this, we applied the long-term time series forecasting-Linear (LTSF-Linear) model, which demonstrates excellent performance in time-series forecasting. Additionally, we integrated feature engineering based on Complete Ensemble Empirical Mode Decomposition (CEEMD) to enhance prediction accuracy. Through this approach, we developed a model capable of delivering high performance in short-, mid-, and long-term forecasting.