Novel performance prediction model of a biofilm system treating domestic wastewater based on stacked denoising auto-encoders deep learning network

Abstract

Stacked denoising auto-encoders (SDAE) deep learning network was used to predict the performance of a two-stage biofilm system, which was constructed based on traditional anaerobic/oxic process. Eight input variables were adopted for performance predicting, including concentrations of chemical oxygen demand (COD), ammonia (NH4+-N) and total nitrogen (TN) of biofilm system influent, concentrations of COD, NH4+-N and TN of anoxic biofilm reactor effluent, influent flow and reflux ratio of biofilm system. While concentrations of COD, NH4+-N and TN of biofilm system effluent were employed as output variables for COD, NH4+-N and TN prediction model, respectively. Root mean square error, mean absolute error, mean relative error and residuals were adopted for evaluating the fitness of the SDAE deep learning network model. Backpropagation neural network, support vector regression, extreme learning machine, gradient boosting decision tree and stacked auto-encoders were adopted for comparison to further demonstrate the effectiveness of the proposed method. Compared with the five contrast models, SDAE deep learning network model showed the best results, suggesting the possible application of performance prediction of the biofilm process with SDAE deep learning network model.