Genetic programming for turbidity prediction: hourly and monthly scenarios

This paper presents the calibration and evaluation of two genetic programming (GP) methods, namely classis GP and gene expression programming (GEP) for turbidity prediction at drinking water distribution networks. Classic GP first method was used to model turbidity at the main water source of Bihac town (Bosnia and Herzegovina) and GEP second method was used to model turbidity at one of the water monitoring stations of city of Antalya, Turkey. The former various predictive models were built based on the mean monthly turbidity measurements recorded during 2006-2018. Moreover, hourly measurements at Gürkavak Station during low turbidity period were used. The results showed that the modelling of turbidity is a challenging task which required careful data analysis especially in the context of determining the optimum lag times/input parameters. No meaningful relation between discharge and turbidity was found at Antalya water supply pipeline. The results also indicated that the predictive models based on the presented algorithms may provide more accurate estimations in comparison to the traditional regression approach. The findings are useful for sustainable urban water management whereby a high quality water supply is aimed.