Dynamic optimization of volatile fatty acids to enrich biohydrogen production using a deep learning neural network

A new strategy was developed to investigate the effect of volatile fatty acids (VFAs) on optimizing bio-H-2 production and improving the efficiency of biogas production. The seed inoculum used was pretreated with four different pretreatments. The relationship between VFAs and biogas compounds was studied as time-dependent components. In time-dependent processes with small sample size data, regression models may not be good enough at estimating responses. Therefore, a deep learning neural network (DNN) model was developed to estimate biogas compounds based on the VFAs. The DNN model was more accurate than regression models and could predict the effect of time changes on biogas compounds. Analysis showed that all the pretreatments were able to increase the ratio of butyric acid/acetic acid (B/A) successfully, decrease propionic acid drastically, and increase the efficiency of bio-H-2 production. The highest efficiency of biogas production was achieved with the highest B/A (2.21) and a low amount of propionic acid (36.6 mg/L), obtained for a digestion time of 67.15 h. Optimal concentration ranges of acetic, propionic, and butyric acids were 823.2-1534.3, 36.3-47.4, and 1522-1822 mg/L, respectively, determined for digestion time of 25.23-123.63 h. These values resulted in the optimum production of bio-H-2, N-2, CO2 , and CH4 in concentration ranges of 6.41-26.25, 12.22-43.24, 5.05-25.3, and 0-1.4 mmol/L, respectively. Therefore, in this time-dependent optimization based on the DNN model, the influence of each VFA on biogas compounds was settled. The optimum amounts were more commonplace to be used in anaerobic digestion.