A Comparative Study of Machine Learning Methods for Pyrolysis Yield Prediction

This paper presents a machine learning-based approach for accurately predicting pyrolysis product yields. Methods such as Linear Regression (LR), K-Nearest Neighbors (KNN), Decision Tree (DT), Support Vector Regression (SVR), Random Forest (RF), and Neural Networks (NN) leverage operating conditions and/or ultimate/proximate analysis data, eliminating the need for reaction kinetics. This innovative approach offers a broader range and higher accuracy of feedstock compared to traditional kinetics-based methods. The KNN model demonstrated superior performance, achieving a correlation coefficient greater than 0.998 and an RMSE of 0.64. These findings provide valuable insights for engineers and practitioners, facilitating the efficient design and operation of pyrolysis units.The selectivity exhibited a notable increase from 2.46 to 5.27. This improvement in selectivity can primarily be attributed to the significantly higher increase in the solubility coefficient of CO2 compared to that of CH4.