Data-Driven model for feedstock blending optimization of anaerobic co-digestion by BMP maximization

Anaerobic Digestion represents an economically and environmentally friendly technology that allows the production of biogas starting from organic substrates. Single substrate digestion often unexploited the total biomass capacity, resulting in low methane yield. On the other hand, it has been proved that biogas production can be significantly improved by combining two or more substrates, performing a co-digestion, which exploits the synergy between different bacteria populations. In the last years, many experimental studies have been conducted to understand how feedstocks interact with each other when mixed, revealing how powerful the blending of substrates could improve its key properties reaching higher methane yield and increasing waste valorization. However, these tests are often time-consuming and rely on the quality of the instruments used to analyze the substrates. Some co-digestion models have been also proposed but are very specific to the co-digestion of a well-defined mix of feedstocks. Unfortunately, ready technologies and models which evaluate an optimal blending ratio, able to estimate the optimal co-digestion configurations, are not discussed so far in the literature. Consequently, this work deals with the development of a tool that can find the optimal blended feedstock composition to produce the highest methane potential. The high number of possible raw materials, and the high variability of their characteristics, reflect the complexity of the problem. So, a database has been created where data about commonly used substrates have been gathered, analyzed, and exploited to build a data-driven model that effectively evaluates the unit’s optimal feedstock composition. Furthermore, the model considers supply-chain issues such as substrate availability and storage options to be more trustworthy in a wider range of industrial settings. Principal influencing parameters of the model were found to be the C/N ratio and the biodegradability of the substrates, while other ones as lipids content and total solid concentration were excluded from the optimization algorithm but still present in the database for further studies. Working temperature of the unit was fixed to 35 ◦C, being the most used condition applied in the literature considered. Finally, the model has been validated by comparing its results to literature experimental tests. from both batch and CSTR (continuous stirred tank) reactors, showing great reliability of the model to the cases analyzed (RMSE <20 mL/gVS for two-substrate mixture, RMSE <30 mL/gVS for three-substrate mixture). Furthermore, the optimization of an industrial case is proposed with data provided by the Th ̈oni company, yielding satisfactory results, such as a total increment of the bio-methane potential up to 70% globally (3-to-5% BMP daily increment) after the diet optimization