(513j) The Hybrid Kinetic-Optimization Modeling and Machine-Learning (ANN and SVM) of Biomass (Maple Leaf) Pyrolysis

Biomass can be utilized to generate heat and valuable chemicals through pyrolysis. Hundreds of chemicals can be produced during the process. Kinetic models involving more reaction steps were generally developed to provide detailed information on the process. However, finding proper kinetic parameters for each reaction step is very time-consuming. In this work, a hybrid kinetic-optimization scheme was proposed to address this issue. A 24-parallel reactions mechanism was developed in the hybrid scheme to describe the thermal decomposition of maple-leaf. A kinetic model using the 24-parallel reaction mechanism was established and coupled with an optimization model to find optimal kinetic parameters of each reaction. The hybrid kinetic-optimization model predicted well-matched results at various conditions. In addition to the kinetic model, the machine learning on biomass pyrolysis was also studied. The artificial neural network (ANN) and support vector machine (SVM) methods were applied to describe pyrolysis of maple-leaf. The impact of important factors for the machine learning methods were investigated. The backpropagation algorithms such as the Levenberg-Marquardt, Bayesian Regularization, Scaled Conjugate Gradient, Adam Stochastic Gradient were examined to ensure that the machine learning models can predict accurate and “unbiased” results. Finally, the performances of the kinetic models and machine learning models were compared and evaluated to explore better modeling options for biomass pyrolysis.