AI based prediction of wastewater treatment plant effluent to supplement the minimal instream flow in the Han River

초록

Securing the minimum instream flow is crucial for utilizing rivers as sustainable water resources and maintaining a resilient ecosystem. For this, the effluent discharged from the J wastewater treatment plant (WWTP) near Hangang Bridge on the Han River (Seoul, South Korea) has been predicted to monitor its contribution to the minimum instream flow using a nonlinear autoregressive exogenous (NARX) model and a support vector regression model using radial basis function kernel (SVR-RBF). Firstly, the discharge flow rate of J WWTP has been predicted based on the influent water quality parameters (i.e., BOD5, COD (or TOC), TN, and TP) and local meteorological data (i.e., humidity and precipitation). Furthermore, parameters were attempted to be more accurately optimized by coupled with principal component analysis (PCA). Simulation without PCA indicated that SVR-RBF outperformed NARX, achieving superior accuracy with RMSE = 1.73%; MAE = 1.23%; and SCC = 0.53. Combining with PCA, both have improved their prediction accuracy higher than without PCA, where SVR-RBF still achieved greater accuracy than NARX. It is decided that the SVR-RBF coupling with PCA can be the most accurate way to predict the WWTP discharge and its influences on the minimum instream flow rate.

키워드