Monthly chlorophyll-a prediction using neuro-genetic algorithm for water quality management in Lakes

Abstract

A genetic algorithm (GA) was combined with artificial neural networks (ANN), designated as neuro-genetic algorithm (NGA) in this study, to determine the effective number of nodes and optimal activated functions (FAs) in an ANN structure. Developed NGA was applied to predict Chlorophyll-a (Chl-a) concentrations in one-month increments in Lakes used as drinking water sources. Correlation analysis was used to setup input parameters. A simulation was conducted for four study sites with the most serious Chl-a problems in South Korea. Results from correlation analysis have indicated that phosphate phosphorus (PO4-P) and electrical conductivity showed high correlation with Chl-a, a factor not often considered in other studies. As the results of prediction of one-month forward Chl-a concentration, NGA showed high accuracy, with averaged determination coefficients of 0.89 and 0.84 in training and testing period, respectively. Double hidden layers showed better performance than a single hidden layer, while a logistic sigmoid function was frequently selected by the genetic algorithm in hidden layers in comparison with linear and hyperbolic tangent function. Practical uses for NGA in proactive water quality management are also discussed in this study.