Forecasting pollutants transport in a river by a fuzzy rule-based model

A fuzzy rule-based model for forecasting polutant transport in a water course is developed. The pollution may be due to accidental causes, such as failure of a wastewater treatment plant or an accidental discharge of pollutant. The model is to provide a forecast of pollutant concentration for decision-making about risk mitigating measures. This problem has generally been studied by solving numerically the classical advective–diffusive transport equation. However, this approach can lead to erroneous solutions either because of the inevitable presence of numerical propagation of errors, or because of the uncertainty in assigning a credible value to the dispersion coefficient that characterizes the diffusive term. Consequently, a fuzzy rule-based model is constructed as an alternative to the classical approach. Simulation of the phenomenon in a well identified case yields data which are then split into a calibration or training set and a validation set. The physical case considered herein is characterized by a constant velocity in the water course and an accidental discharge of pollutant represented by a step function. The training data set is applied to calibrate the rules. The model reliability is investigated by comparing fuzzy model output values with validation data set values. Results are encouraging, so that the fuzzy rule-based model may be used for forecasting on the basis of experimental data even in the case when no partial differential equation (or solution thereof) is available.