Integrating operating conditions to develop a neural network for predicting organics removal and power density in an earthen microbial fuel cell treating leachate

Nowadays, the acceptability of artificial neural networks (ANNs) to develop models for biological treatment systems has been broadened instead of the complex mechanistic mathematical model. The present study attempts to train neural networks to predict organics removal and power output from an earthen microbial fuel cell treating acidogenic leachate obtained from kitchen waste. To model the effluent chemical oxygen demand (COD) and volatile fatty acid (VFA) values and power density, the study used four input parameters: influent COD, and VFA, hydraulic retention time, and anode electrode surface area. Among the multiple algorithms employed, the Bayesian inference proved to deliver a winning model to forecast the response variable with an overall high coefficient of determination of 0.9890. This high correlation for a biological system employing a real substrate, i.e. leachate, ensures the model's reproducibility and adequacy. Additionally, the statistical error criteria for model efficacy showed acceptable values. Besides, the study performs validation runs with an error of approx. 11% with conservative predictions. Ultimately, the ANN model fared well in predicting the response for effluent COD and power output when compared with kinetic study (Han-Levenspiel) and statistical equation (from ANOVA) to the experimental values.