Predicting real-time coagulant dosage in water treatment by artificial neural networks and adaptive network-based fuzzy inference system

Abstract

Coagulation is an important component of water treatment. Determining the optimal coagulant dosage is vital, as insufficient dosage will result in unqualified water quality. Traditionally, jar tests and operators’ own experience are used to determine the optimum coagulant dosage. However, jar tests are time-consuming and less adaptive to changes in raw water quality in real time. When an unusual condition occurs, such as a heavy rain, the storm water brings high turbidity to water source, and the treated effluent quality may be inferior to drinking water quality standards, because the conventional operation method can be hardly in time to adjust to the proper dosage. An optimal modeling can be used to overcome these limitations. In this paper, artificial neural network (ANN) and adaptive network-based fuzzy inference system (ANFIS) models were used to model poly aluminum chloride (PAC) dosing of northern Taiwan's surface water. Each of them was built based on 819 sets of process-controlled data. The performance of the models was found to be sufficient. Two simulation tools, ANN and ANFIS, were developed that enabled operators to obtain real-time PAC dosage more easily. The self-predicting model of ANFIS is better than ANN for PAC dosage predictions.