Comparative analysis of selected plastic pyrolysis and catalytic versus non-catalytic pyrolysis of polyethylene using artificial neural networks for oil production.

Research on plastic pollution is crucial, particularly with the recent emphasis on converting waste plastics into oil for sustainable energy. Very few studies have utilized artificial neural network (ANN) modeling for plastic thermal conversion, such as predicting fuel yield from mixed plastics and performing sensitivity analyses to identify which plastics produce more oil. Meanwhile, no study has conducted a comparative analysis of different models for catalytic and non-catalytic thermal conversion of various plastics, nor has a sensitivity analysis of process parameters using ANN for oil production. This study aims to (1) validate and predict oil yield across different catalytic and non-catalytic thermal conversion processes for plastics using MATLAB-based ANN training; (2) perform sensitivity analysis on process parameters affecting oil production; and (3) forecast oil yield using virtual input parameters not included in real experiments. The models demonstrate R values near 1 and mean squared error (MSE) values close to zero, indicating strong validation. For catalytic polyethylene (PE) pyrolysis, the impact ranking is reaction temperature (36.9 %) > pressure (32.1 %) > Zn loading in ZSM5 (30.9 %). In non-catalytic PE and biomass co-torrefaction, the impact ranking is reaction temperature (47.2 %) > feedstock-to-solvent ratio (23.9 %) > biomass-to-PE ratio (16.6 %) > experimental duration (12.1 %). For catalytic mixed plastic (MP) torrefaction, the ranking is reaction temperature (54.8 %) > duration (18.4 %) > solid-to-liquid ratio (15.9 %) > NaOH amount (10.8 %). In non-catalytic MP pyrolysis, the significance ranking is particle size (44.51 %) > pyrolysis temperature (34.4 %) > pyrolysis duration (21.06 %). Accordingly, temperature, catalyst loading, and duration are critical for catalytic processes, while particle size and temperatures affect non-catalytic pyrolysis. The predicted and experimental outcomes differ by only 1 to 3, demonstrating that the models accurately simulate the predicted values. This study uses ANN sensitivity analysis to compare catalytic and non-catalytic methods, offering insights into scale-up applications and sustainability.