Archives

  • Archives
  • >
  • Home

DOI : ,    Vol.10, No.4, 51 ~ 61, 2002
Title
Application of Neural Network for Downstream Water Quality Prediction in Dry Period
김주환 Ju Hwan Kim
Abstract
In dry period, the water quality deterioration demands additional outflow from dam due to the exhausted water in branch of almost rivers. Some extent of deterioration of water quality can be attenuated by dam reservoir operation to control outflow considering predicted water quality at downstream in dry season. Multiple regression and neural network models are developed in this study to predict the monthly concentration of NH_3-N in downstream by the variation of dam outflow. The multiple regression and neural network models have advantages as followings : once the models are constructed by using enormous historical records, those models can be used directly for prediction without various input data which is inevitably needed in mathematical models as boundary conditions. The autoregressive characteristics of NH_3-N are analyzed by auto-correlation coefficient function for determining monthly lag time. Monthly temperature, outflow and alkalinity data are considered as model variables. Three models are developed by using 72 monthly data between Jan. 1993 and Dec. 1998 and 24 month of data between Jan. 1999 and Dec. 2000 are used for verification. The model performances are evaluated by determination coefficients between the observed and the predicted water quality. The determination coefficient values of three models are more than 0.92. The results show that neural network model can be applied well to predict water quality in stream river and give more excellent results than multiple regression model.
Key Words
pdf view PDF