Neural networks modelling of pesticides removal by activated carbon for water treatment

Abstract

The adsorption equilibrium of pesticides (atrazin, atrazin-desethyl and triflusulfuron-methyl) onto activated carbon (AC) carried out in batch reactors has been determined for a large range of concentrations (from 5 μg/L to 21.4 mg/L) to characterise adsorption mechanisms. Single-solute isotherms tend to confirm the decisive role of the adsorbent's microporosity in the adsorption capacity of the AC. These adsorption capacities are high and range between 63 and 509 mg/g. The adsorption of the three pesticides is also studied in a dynamic reactor. The influence of operating conditions (initial concentration Co, flow velocity Uo) and adsorbent's characteristics is investigated. All dynamic experimental results are modelled by a neural network to establish the link between the characteristics of activated carbon materials and the adsorption's results. Parameters related to the adsorbate–adsorbent affinity in a batch reactor are consequently introduced in the input layer of the neural network added to operating conditions whose influence was shown (Co and Uo) and time t. The statistical quality of the neural network modelling is high (R2=0.985 for the static neural network and R2=0.993 for the dynamic neural network between experimental and predicted values for the test data set).