Application of machine learning to pyrolysis reaction networks: Reducing model solution time to enable process optimization. (2nd September 2017)

• Record Type: Journal Article
• Authors: Hough, Blake R.; Beck, David A.C.; Schwartz, Daniel T.; Pfaendtner, Jim
• Is Part Of: Computers & chemical engineering. Volume 104(2017)
• Page Start: 56
• Page End: 63
• Publication Date: 2017-09-02
• Subjects: Neural network -- Kinetic modeling -- Pyrolysis -- Biomass; Chemical engineering -- Data processing -- Periodicals; 660.0285
• DOI: 10.1016/j.compchemeng.2017.04.012
• Languages: English
• ISSNs: 0098-1354

Abstract

Highlights: Neural networks are an appropriate modeling tool to represent high dimensional detailed kinetic models with high accuracy and reduced computational cost. The speed of a neural network kinetic model facilitates using detailed kinetics process models with complex spatiotemporal heterogeneity. The information content contained within the neural network is tunable depending on the desired outputs. Abstract: Comprehensive models of biomass pyrolysis are needed to develop renewable fuels and chemicals from biomass. Unfortunately, the detailed kinetic schemes required to optimize industrial biomass pyrolysis processes are too computationally expensive to include in models that account for both kinetics and transport within reacting particles. Here we present a machine learning approach using artificial neural networks and decision trees to reduce the computational expense of detailed kinetic models by four orders of magnitude, enabling their use in comprehensive models. The trained neural networks generalize very well, predicting the outputs of the detailed kinetic model with over 99.9% accuracy on new data. The machine learning approach we outline is not specific to kinetic modeling and can be applied to any set of input and output data, even if the underlying relationship between inputs and outputs is unknown.