Correspondence of the Temporal Variation of Actual STP over Predicted STP discharge to Actual Instream Flow

초록

Instream flow is the minimum flow rate that must flow through a river to maintain its normal function for necessalily sustainable ecosystem. The ratio of flows of sewage treatment plant (STP) effluent to the instream of the Hangang Birdge in Han river was varied at 36.2% on average and 49.2% for the maximum. In this regard, it was supposed that the instream flow can be closely related with that of STP effluent discharge so that the STP effluent discharge flow would be accordingly predicted in this study. For the first, the effluent discharge flows from two STPs of Jungrang and Tancheon, were predicted by using two different sets of data(daily collected for 6yrs); STP influent water quality parameters of BOD, COD, TN, and TP and meteorological condition data daily logged for 6yrs for 2016 to 2021 nearby in the STP such as precipitation, humidity, and temperature. These data were thus pre-processed and reduced to two dimensions through Principal Component Analysis (PCA), thereafter a biplot was used to identify the most important set of variables. Using those data, the prediction was performed by the Nonlinear Autoregressive Exogenous Model (NARX) neural network provided by MATLAB. where 80% and 20% of data have been subsequently used for neural network training and testing, respectively. Among the neural network training algorithms, the SCG algorithm revealed the best fit among other showing that RMSE of the median(Root Mean Square Error) on the ratio of temporal variation of the actual STP over the predicted to the instre flow was 1.32×105 m3/day. Nevertheless, it will be further necessarily ensued to more dwindle the level of MAPE, RMSE for accquiring the more enhanced prediction accuracy for the STP discharge by changing the number of hidden neurons and the time delay by using different neural networks of MLR, SR.