Feed-Forward Neural Network Approximation Applied to Activated Sludge System

The dynamic behavior of an activated sludge system is highly complex and uncertain. To efficiently control and operate the system, a reliable model capable of accurately describing the several time-varying parallel processes of the system is needed. Most of the existing models are too complex to use for design or control purposes. This paper presents a feed-forward neural network model for the system. The model validation was achieved through the use of appropriate international accepted data of the benchmark simulation model no. 1 (BSM1). Simulation studies revealed that the neural network model exhibited an outstanding performance in predicting the effluent quality, root mean square error (RMSE) of 0.0464, mean absolute deviation (MAD) of 0.0347, correlation coefficient (R) of 0.979 for chemical oxygen demand (COD) and RMSE of 0.1103, MAD of 0.0794, R of 0.841 for the total nitrogen (TN) could be acheived. The model is quite effective and suitable tool for the activated sludge system.