(cest2017_01138) Comparison of Artificial Neural Network and State Space Model for Predicting River Water Quality

摘要

The purpose of this study is to investigateappropriate tools for river water quality forecasting asvariations in water quality are difficult to predict due to thecomplicated nature within the range of various waterquality factors. In this study, state space and neuralnetwork models are employed to mathematically analyzethe intricate nonlinearity of processes that affect factorsrelated to water quality. A monthly forecasting model isproposed that can predict water quality parameters,including dissolved oxygen (DO), biochemical oxygendemand (BOD), and suspended solid (SS) at the Miho riverstation in the Geum river basin (Korea). River waterquality is predicted through the learning and theverification processes after applying the neural networktheory to the proposed water quality forecasting model.Practical applications for predicting water qualityprediction are examined by comparing the proposed modelto the state space model (SSM). As a result, the artificialneural network (ANN) is estimated to have the ability topredict water quality more accurately than the state spacemodel for each water quality item.