Simulation of a coke wastewater nitrification process using a feed-forward neuronal net

A laboratory-scale Activated Sludge System (ASS) was employed for the biodegradation of coke wastewater, which contains high concentrations of ammonium, thiocyanate, phenols and other organic compounds. The well-known kinetics models of Monod or Haldane are not very useful due to inhibition phenomena amongst the pollutants and also, they need the determination of a wide range of parameters to be introduced in the models. In this paper, a feed-forward neural network is outlined to obtain a satisfactory approach for estimating the effluent ammonium concentration of the treatment plant. The methodology consists in performing several tests with a group of different sizes of the hidden layer and different subsets of input variables. The developed model is useful to obtain simulations under different conditions of the influent stream, thus enabling the effluent ammonium concentration to be estimated. This neural network achieves better results than classical mathematical models for biological wastewater treatment as a result of the complex composition of the coke wastewater.