Application of Data-Driven and Optimization Methods in Identification of Location and Quantity of Pollutants

Abstract

Water pollution is one of the major problems in providing and preserving water resources, so identifying the pollution source plays a critical role in regulation actions. Thus, this paper addresses the process of pollution source identification, including location, concentration, and the time of injection in surface water by using a data-mining method [artificial neural network (ANN)] and optimization techniques [genetic algorithm (GA) and pattern search (PS)]. The CE-QUAL-W2 numerical model is used to produce input and output data in ANN and simulation models. To check the capability of the methodology, the identification of various hypothetical examples of pollution with several forms of injection of the pollutant in a nonprismatic water canal is performed. Results of data-driven and optimization methods are evaluated by employing statistical criteria. Final results show that the ANN method is capable of identifying a pollutant injection hydrograph and it is relatively sensitive to the accuracy of monitoring so that like the optimization method for errorless data, the determination coefficient (