Derivation of Kinetic Parameters and Lignocellulosic Composition From Thermogram of Biomass Pyrolysis Using Convolutional Neural Network

A novel method employing a 1-dimensional convolutional neural network (1D-CNN) has been developed to deduce kinetic parameters for the three-parallel-reaction model (TPRM) and the lignocellulosic composition from the thermogram of biomass pyrolysis. This model was trained on differential thermogram (DTG) datasets created at various heating rates with rate constants randomly selected from expansive ranges. Furthermore, to enhance prediction accuracy, a denoising autoencoder (DAE) was crafted to eliminate noise from experimental data effectively. The 1D-CNN regression model forecasted kinetic parameters with mean errors of 1.52% for trained heating rates and 1.39%-3.19% for other heating rates. When tested on four biomass samples, the model precisely mimicked the DTG curves with R2 values ranging from 0.9956 to 0.9994. Relative to conventional numerical methods, this model delivers comparable prediction accuracy but through a significantly streamlined and expedited process. Enhancements are needed to broaden the model's applicability across various kinetic models and materials.