Machine learning estimation of biodegradable organic matter concentrations in municipal wastewater.

This study investigates whether a novel estimation method based on machine learning can feasibly predict the readily biodegradable chemical oxygen demand (RB-COD) and slowly biodegradable COD (SB-COD) in municipal wastewater from the oxidation-reduction potential (ORP) data of anoxic batch experiments. Anoxic batch experiments were conducted with highly mixed liquor volatile suspended solids under different RB-COD and SB-COD conditions. As the RB-COD increased, the ORP breakpoint appeared earlier, and fermentation occurred in the interior of the activated sludge, even under anoxic conditions. Therefore, the ORP decline rates before and after the breakpoint were significantly correlated with the RB-COD and SB-COD, respectively (p < 0.05). The two biodegradable CODs were estimated separately using six machine learning models: an artificial neural network (ANN), support vector regression (SVR), an ANN-based AdaBoost, a SVR-based AdaBoost, decision tree, and random forest. Against the ORP dataset, the RB-COD and SB-COD estimation correlation coefficients of SVR-based AdaBoost were 0.96 and 0.88, respectively. To identify which ORP data are useful for estimations, the ORP decline rates before and after the breakpoint were separately input as datasets to the estimation methods. All six machine learning models successfully estimated the two biodegradable CODs simultaneously with accuracies of ≥0.80 from only ORP time-series data. Sensitivity analysis using the Shapley additive explanation method demonstrated that the ORP decline rates before and after the breakpoint obviously contributed to the estimation of RB-COD and SB-COD, respectively, indicating that acquiring the ORP data with various decline rates before and after the breakpoint improved the estimations of RB-COD and SB-COD, respectively. This novel estimation method for RB-COD and SB-COD can assist the rapid control of biological wastewater treatment when the biodegradable organic matter concentration dynamically changes in influent wastewater.

A simulation of acetate consumption and electricity generation in a single microbial fuel cell considering the diversity of nonelectrogenic bacteria.

To simulate acetate consumption and electricity generation in a cycle of a microbial fuel cell (MFC) treating synthetic acetate-based wastewater with low concentration, nonelectrogenic bacteria (NEB), which had no contribution in electricity generation, was incorporated with methanogen's kinetic parameters into a previous biofilm model proposed by Marcus et al. (Biotechnol Bioeng 98:1171-1182, 2007). However, the Coulombic efficiency was estimated to be 40.1%, whereas the experiment showed 13.6%, as the presence of NEB was obviously underestimated. Thus, the maximum NEB reaction rate (qmaxC) was temporarily calibrated, and a sensitivity analysis was then conducted. As a result, the growth parameters of NEB, the growth of the exoelectrogenic bacteria, and the biofilm detachment were identified as influential parameters. qmaxC and a half rate constant of NEB (KsC) were selected as potential calibration parameters. The two sets of calibrated parameters (0.342 mmol-acetate (Ac)/mg-volatile solids (VS)/d of qmaxC and 33.8 mg-carbon (C)/L of KsC; 0.274 mmol-Ac/mg-VS/d of qmaxC and 16.9 mg-C/L of KsC) showed a good agreement with the experimental results at 100 mg-C/L of initial acetate. However, the calibrated parameter values obviously differed from those in previous models. The calibrated model also showed good agreement with the experimental results at 50 and 200 mg-C/L of the initial acetate. In view of the different values of qmaxC and KsC from those of methanogenic bacteria in previous models and the previous findings on anode microbial community, which showed that NEB are not only methanogenic bacteria, we concluded that the diversity of NEB should be considered to simulate performances in a cycle of MFC treating low organic matter concentrations.