Abstract:
Artificial intelligence (AI) techniques have been successfully performed in many different
water resources applications such as rainfall-runoff, precipitation, evaporation, discharge (Q),
dissolved oxygen (DO), chemical oxygen demand (COD), biological oxygen demand (BOD),
sediment concentration and lake levels by many researchers over the last three decades. In this
study, three different adaptive neuro-fuzzy inference system (ANFIS) techniques, ANFIS with
fuzzy clustering (ANFIS-FCM), ANFIS with grid partition (ANFIS-GP) and ANFIS with
subtractive clustering (ANFIS-SC), were developed to estimate COD concentration by using
various combinations of daily input important variables water suspended solids (SS), discharge
(Q), temperature (T) and pH. Root mean square error (RMSE), mean absolute error (MAE) and
determination coefficient (R2) statistics were used for the comparison criteria. Training, testing
and validation phase’s results of the optimal ANFIS models were also graphically compared
each other. Comparison of the results indicated that the ANFIS-SC(1,0.3,1) model whose input
is water SS was found to be slightly better than the other models in estimation of COD
according to the comparison criteria in testing phase. In the validation phase, however, ANFISFCM(
1,3,gauss,1) model performed slightly better than ANFIS-GP(3,trimf,constant,1) and
ANFIS-SC(1,0.3,1) models. It can be said that three different ANFIS techniques provide similar
accuracy in estimating COD.
