Neuro-fuzzy ensemble techniques for the prediction of turbidity in water treatment plant

Providing a satisfactory and reliable prediction tool for Turbidity in water treatment plant is quite an essential task for various environmental and public health perspective. In this paper, a neuro-fuzzy approach is developed using two different optimizations of fuzzy inference system (FIS) (i.e. hybrid and backpropagation) to predict the treated Turbidity at Tamburawa water treatment plant (TWTP). Subsequently, a neuro-fuzzy ensemble technique was applied to improve the performance of the two optimizations. For this purpose, the daily recorded data of turbidity (Turb
_R
) (μs/cm), conductivity (Cond
_T
) (mS/cm), total dissolve solid (TDS
_R
) (mg/L), chloride (mg/L) and suspended solid (SS
_R)
(mg/L) and Hardness (Hardness
_T
) (mg/L) from TWTP were obtained. The predictive models were evaluated based on two numerical indicators (determination coefficient and root mean square error). The obtained results indicated that neuro-fuzzy hybrid increased the performance accuracy of neuro-fuzzy backpropagation optimization up 16% and 15% in both the training and testing phase respectively. For neuro-fuzzy ensemble results, the performance proved that hybrid ensemble increased the prediction efficiency of backpropagation ensemble up to 18% in the testing phase. Hence, for the prediction of Turbidity in TWTP both the hybrid FIS optimization and ensemble hybrid FIS optimization showed excellent accuracy while for its recommended to employed ensemble techniques in case of backpropagation FIS optimization. The ensemble methodology proved to be implemented as a real-time prediction model that can provide a brilliant approach for environmental sustainability.