Artificial neural networks for bio-based chemical production or biorefining: A review

Machine learning through artificial neural networks have emerged as vital tools to predict chemical behavior for many of the most recognized biomass valorization processes relevant to biorefineries for the purpose of opti-mization of desired products and reaction conditions. Until recently, these neural network methodologies have successfully been utilized in the petroleum industry where much more extensive databases are available for effective algorithm training. These systems provide compelling advantages for pattern recognition when inter-preting the influence of ever-changing feedstock compositions for complex biomass conversion processes as they do not require any a prior knowledge of reaction mechanisms or thermodynamic phenomena. This has been revealed to be tremendously beneficial for real-time, dynamic control applications of biochemical processes for rapid parameter monitoring and regulation such as during fermentation or anaerobic digestion. This review aims to present and evaluate studies that have attempted to apply these neural network strategies to various aspects of biorefining and how these models address the common challenges that occur when relying on conventional mechanistic modelling approaches to estimate sophisticated, non-linear systems. Comparisons are then identified when implementing these artificial intelligence computing practices in traditional petroleum refineries where feedstock inconsistencies are not as paramount compared to biorefineries. Subsequently, the practicality of these neural networks is critically assessed and recommendations are presented on how to strengthen its applicability and predictability towards future bio-based chemical production. Mathematical models such as artificial neural networks will be an integral technology in the future bioeconomy for the realization of innovative biorefinery concepts as computational power continues to advance.

Automated summary

Artificial neural networks through machine learning are vital tools to predict chemical behavior and optimize biomass utilization processes. They offer advantages in dynamic control applications and address challenges of conventional modeling methods. Their practicality and predictability towards bio-based chemical production are critically assessed for future advancements in the bioeconomy.