DEVELOPING A WATER QUALITY ESTIMATION MODEL BY INTEGRATING DEEP LEARNING WITH NONLINEAR TIME-SERIES ANALYSIS

Abstract

 This study developed an autoregression-driven deep neural network model using deep learning techniques and nonlinear time-series analysis to estimate water quality variations in coastal areas. This local prediction model analyzes the autoregression characteristics of the nonlinear water quality system combined with an extrinsic deep learning model to express the relationship between water quality items and external factors such as tides and weather.

 By combining both models, two water quality items were estimated: the electrical conductivity in a tidal river and the dissolved oxygen concentration in the bottom layer of an enclosed bay. Both models showed high accuracy in estimating the two water quality items. However, the autoregression-driven deep learning model was superior, particularly for water quality items affected by several internal state variables, such as physical, biological, and chemical processes.