The Utility of Inductive Models using Artificial Intelligence (AI) Techniques for Water Quality Modeling in a TMDL Framework

This paper describes the utility of inductive models developed using two Artificial Intelligence (AI) techniques for water quality modeling in a TMDL framework. The two AI techniques used in the development of inductive models include Artificial Neural Networks (ANN) and Fixed Functional Set Genetic Algorithms (FFSGA). Inductive models are becoming more popular these days due to their ease of use and simplicity as a substitute for the more process-based deductive models for water quantity and quality modeling. Such models can be very effective in making informed and timely decisions for watershed management, specifically in the real time control of water resources systems. In this paper, inductive models using these two techniques are developed for modeling fecal coliform concentrations in surface water based on real time stream flow and water quality monitoring of streams. The resulting models are used to predict fecal coliform concentration for given site based on constituents such as stream flow and turbidity. The model performance of these models is evaluated, by comparison to actual fecal concentrations monitored for the site, both in training and validation data sets and compared to actual fecal concentrations monitored for the site.