Maximising Liquid Fuel Production From Waste Polystyrene Pyrolysis: Comparing Response Surface Methodology and Artificial Neural Network Modelling

ABSTRACT

Waste polystyrene is a major environmental problem due to its high production rate and resistance to natural breakdown. This study examines converting waste polystyrene into liquid fuel through batch pyrolysis, focusing on process optimisation and predictive modelling. The feedstock was characterised using proximate, ultimate and thermogravimetric analyses, confirming high volatile content and favourable thermal degradation suitable for pyrolysis. Response Surface Methodology with a Central Composite Design was used to study the effects of feedstock weight, particle size, reaction temperature and residence time on liquid fuel yield. An Artificial Neural Network model was separately developed to predict fuel yield across varying operating conditions. The RSM model identified optimal conditions at 567.7 g feed weight, 490°C, 86.7 min residence time and 2.0 mm particle size, giving a maximum liquid fuel yield of 78.09% with strong statistical agreement. The ANN model, trained using multiple algorithms, showed high predictive accuracy, with the Scaled Conjugate Gradient algorithm performing best (overall R = 0.931; testing R ²  = 0.960, MSE = 4.87, RMSE = 2.21, MAE = 1.93). The RSM model achieved a higher R ² of 0.9999 and a lower RMSE of 0.0746, confirming better fitting accuracy, while ANN predictions remained reliable. Fuel characterisation using FTIR and GC‐MS confirmed aliphatic and aromatic hydrocarbons, with density and calorific value comparable to commercial liquid fuels. Overall, the results show RSM offers stronger optimisation capability while ANN provides dependable yield prediction, highlighting the combined value of statistical and machine learning methods for waste polystyrene valorisation.