Fuel composition forecasting for waste tires pyrolysis process based on machine learning methods

Waste tire pyrolysis oil holds significant potential as an energy source and chemical feedstock, playing a crucial role in enhancing resource efficiency and promoting environmental sustainability. This study aims to develop a reliable forecasting model for determining the composition proportions of waste tire pyrolysis oil and assessing its performance. This study employed the gradient boosting decision tree (GBDT) method to develop the model that accurately forecasts the composition proportions of aromatic, aliphatic, and other compounds present in waste tire pyrolysis oil. Importance in XGboost and Cramer's V analysis in correlation analysis were used to select the input variables for the forecasting mode. Additionally, this study normalized the percentages of the three components to ensure consistency. To evaluate the performance of the proposed models, we compared them with 13 models from the Decision Tree series, BPNN series, SVM series, and HDMR series, thereby verifying the accuracy of our models. Furthermore, this study conducted random sampling to examine the stability and generalization ability of our proposed models. The experimental results demonstrate that the forecasting model exhibits excellent performance in estimating the proportions of aromatic compounds, aliphatic compounds, and other compounds. Over 60% of the samples yielded relative error percentage (REP) values below 30%. This indicates that this proposed model is capable of accurately forecasting the composition proportions of waste tire pyrolysis oil. The proposed model holds the potential to facilitate dynamic optimization of the waste tire resource reuse process and contribute to the sustainable development of waste tires.