Modeling of biogas production from food, fruits and vegetables wastes using artificial neural network (ANN)

•Cumulative biogas production was studied for food, fruits and vegetables wastes.•Database was built on mixed values of eight variables seen in literature.•Different artificial neural network (ANN) topologies were implemented and assessed.•The predictions showed more than 85% correctness using the complete database. Biogas can be generated from many types of biomass residues. These biomasses have different characteristics which makes standardization of process conditions difficult. Thus, the influence of different conditions has been explored for several biogas production scenarios using numerical models. In the present work, an experimental study of biogas production from food waste was carried out in triplicate in a batch reactor at 37 °C with an organic loading rate (OLR) equal to 5, 10 and 20 g VS/(l.d) after 21 days. A database was also built using values presented in the literature in order to develop a numerical model using artificial neural networks (ANN), for food waste (FW), fruit and vegetables waste (FVW) or blends of both in codigestion (CD), reactor/feed type, volatile solid (VS), pH, OLR, hydraulic retention time, temperature and reactor volume, as input variables, and the cumulative biogas production as output. The response surfaces of the ANN model were found to be useful for defining the optimum region in biogas production; when applied to the training, testing and validation datasets, the model showed acceptable values of coefficient of determination (0.9929, 0.8486 and 0.6167 for the input parameters respectively). It was found that the production of biogas was the highest when temperatures was within the range of thermophilic conditions, with a local maximum for mesophilic conditions. Optimized biodigestion of CD or FW allows higher VS content (around 15–20%) than for FVW (lower than 10%). It was also observed that biodigestion of FVW leads to the highest cumulative biogas production (around twice the value found for FW and CD).