Prototype Artificial Neural Network for Prediction of Arsenic Breakthrough Curves from Full-scale Small Drinking Water Treatment System Data

摘要

In 2006, the Maximum Contaminant Level (MCL) for arsenic was reduced from 50 μg/L to 10 μg/L under the NationalrnPrimary Drinking Water Regulations (NPDWR). The regulatory change prompted an investigation of treatmentrnalternatives for meeting the new requirements and significant amounts of data were generated from bench-scale, pilot, andrnfull-scale demonstration projects. Data from demonstration projects were imported into a prototype artificial neuralrnnetwork (ANN) model to predict arsenic breakthrough curves from full-scale small drinking water treatment facilities.rnData on contaminant concentration, water quality parameters, operational conditions and observed system performancernwere compiled from 16 full-scale demonstrations employing absorptive media for arsenic removal that reflected differentrngeographical, hydrological, water quality, operational configurations and hydraulic loading conditions. A four-layerrnperception model was developed to predict arsenic breakthrough patterns based on specific water quality and operationalrnconditions. The model was validated by comparing predictions with the performance of similar treatment scenarios.rnFuture applications of this type of analysis include decision-support tools linking water quality, operational andrnperformance data to screen treatment technologies and assist utilities in solving simultaneous compliance issues.