Wastewater BOD Forecasting Model for Optimal Operation Using Robust Time-Delay Neural Network

Due to the lack of reliable on-line sensors to measure water quality parameters, it is difficult to control and operational optimization in the wastewater treatment plants (WWTPs). A hybrid time-delay neural network (TDNN) modeling method with data pretreatment is applied for BOD forecasting model in wastewater treatment. PCA is combined with robust expectation-maximization (EM), which reduces the influence of noise, outliers and missing data. The principal components are used as inputs to time-delay neural networks to predict the effluent BOD value. The simulation results using real process data show an enhancement in speed and accuracy, compared with a back propagation neural networks.