Current density performance prediction for a microbial electrolysis cell using machine learning methods

Accurate prediction of microbial electrolysis cell (MEC) performance is a key step towards the reliable operation and industrial deployment of this technology. However, its modelling and prediction are complex due to the nonlinear nature of the physical, chemical and biochemical processes involved, as well as its technical configuration. This study proposes an advanced machine-learning-based modelling framework to predict the bioanode current density in an MEC system using experimental data. Eight machine learning models were evaluated: Random Forest (RF), XGBoost, multilayer perceptron (MLP), convolutional neural network (CNN), long short-term memory (LSTM), gated recurrent unit (GRU), nonlinear autoregressive neural network with exogenous variables (NARX), and a hybrid CNN-NARX model. Through experimentation, data were collected on pH, conductivity, optical density, chemical oxygen demand, acetic acid, butyrate, and propionic acid concentrations, as well as current density, in a MEC. The results demonstrated that NARX and CNN-NARX models achieved superior predictive performance. Notably, the CNN-NARX model attained the highest coefficient of determination (R 2 = 0.87), whereas the NARX model recorded the lowest prediction errors (MSE = 0.000082, MAE = 0.008047, and RMSE = 0.009029). These findings suggest that modelling architectures capable of capturing the intricate nonlinear dynamics of the MEC can effectively simulate its bioelectrochemical response. The SHAP analysis identified pH, acetic acid, and optical density as the most influential predictors, cumulatively accounting for 65.5% of the overall feature importance. The results offer an empirical proof of concept for the application of data-driven models in predicting current density and bioelectrochemical monitoring within MEC-based wastewater processes.