| Abstrakt: |
Marabú (Dichrostachys cinerea), a fast-growing shrub species, has garnered interest as a potential energy crop due to its properties. In developing thermochemical processes for utilising D. cinerea, specifically through pyrolysis, precise prediction of its behaviour is essential for optimising process efficiency and understanding the underlying mechanisms. This study focuses on comparing the effectiveness of kinetic and artificial neural network (ANN) modelling methods in predicting the pyrolysis of D. cinerea. Utilising thermogravimetric data at four different heating rates (5, 10, 20 and 40 °C/min), a kinetic model based on three independent parallel reactions was developed. In the ANN model, the input variables (heating rate (°C/min), temperature (°C) and time (min)) were used to predict the output variable: weight loss (%). To optimise a backpropagation neural network (BPNN), 4-fold cross-validation and Bayesian optimisation were employed. The findings demonstrate that both methods effectively predict weight loss, with the ANN model achieving superior accuracy in capturing experimental data, particularly at local maxima of weight loss, reflected by R2values exceeding 0.99. The ANN method excels without the need for predetermined kinetic reaction mechanisms, showcasing its ability to adapt to complex, non-linear types of behaviour more accurately than traditional models. This study not only provides valuable insights into the pyrolytic behaviour of D. cinereabut also establishes a benchmark for future research in the predictive modelling of pyrolysis for diverse types of lignocellulosic biomass. |
Full text from SpringerLink