Pine needle pyrolysis by thermogravimetry: comparison between kinetic analysis and artificial neural network

The pine needle (PN) pyrolysis is investigated by thermogravimetry (TG) under nitrogen. Six kinetic models and 19 artificial neural networks (ANN) are compared based on the number of adjustable parameters. A 5-step kinetic model, assuming parallel reactions of moisture, extractives, hemicellulose, cellulose, and lignin, is proposed for PN pyrolysis. The ANN-2-4 (the first hidden layer has two neurons, and the second hidden layer has four neurons) is the best ANN model to predict PN pyrolysis. Furthermore, two model-fitting methods, EA method and ETp method (Zhu and Liu in Thermochimica Acta 690:178686, 2020. https://doi.org/10.1016/j.tca.2020.178686), and two optimization algorithms, genetic algorithm and trust region reflective algorithm (TRRA), are applied for kinetic analysis. The Levenberg-Marquardt algorithm is adopted to optimize adjustable parameters for ANN model. Previous work pointed out that a simple optimization algorithm achieves faster calculation, but is not suitable for kinetic analysis of complex chemistry. However, the results show that the ETp method combined with TRRA (a simple optimization algorithm) performs better than other kinetic analysis methods for analyzing PN pyrolysis. This work is the first effort to estimate kinetic parameters by the ETp-TRRA method.