Optimization of dark fermentation for biohydrogen production using a hybrid artificial neural network (ANN) and response surface methodology (RSM) approach

Herein, the production of biohydrogen by dark fermentation was optimized using a novel hybrid approach that combines ANNs (artificial neural networks) with RSM (response surface methodology). Using the limited numbers of data (15 runs) as training data set together with one cross-out method for validation, the complete 29 runs of well-established data matrix were created from ANNs for RSM statistical analysis in order to correlated the critical process parameters with hydrogen production performance. This methodology was found to be robust, cost-effective, reliable, and can be extensively analyzed the critical operational parameters, that is, carbon sources (obtained from potato peel wastes and starchy wastes), metal cofactor Fe0, pH, and dose levels of microbes on the hydrogen production, along with concentrations of other metabolites, such as acetic acid, propionic acid, butyric acid, valeric acid, and ethanol. The established ANNs-RSM model using Box–Behnken design indicates the significant changes caused by the variations of a few critical operation parameters. The resultant model shows an exceptionally good result in terms of nonlinear noisy processes. Both single and multiple objective optimizations for dark hydrogen fermentation can achieve by using the established hybrid ANN-RSM system. The optimal operating conditions (starch 6.2 kg/m3, pH 6.7, Fe0 11.7 g/m3, sludge 24.6 g/m3) could lead to the generation of hydrogen with a yield of 106.2 (cm3/g) and metabolites, that is, propionic acid (2.8 kg/m3), butyric acid (2E−2 kg/m3), valeric acid (4E−4 kg/m3) acetic acid (1.9 kg/m3), and ethanol (0.1 kg/m3) simultaneously