Machine learning-driven prediction and optimization of pyrolysis oil and limonene production from waste tires

Abstract

The pyrolysis oil from waste tires possesses significant economic value and is a crucial factor in determining the industrial viability of tire pyrolysis processes. Limonene in pyrolysis oil is a significant component with extremely high industrial application value. In the process of tire pyrolysis, predicting the pyrolysis products through operating conditions and feedstock composition can effectively control industrial operations and enhance operational efficiency. However, there is currently a lack of robust prediction methods for pyrolysis oil and limonene yield. This study proposes the application of machine learning to predict the yield of tire pyrolysis oil and limonene. Artificial Neural Network (ANN) and Random Forest (RF) models were developed to create prediction models. In the statistical analysis, RF achieved optimal R2 median values of 0.83 and 0.64 for pyrolysis oil and limonene predictions during the testing stage.For the prediction of limonene yield, the best R2 and RMSE values in the testing and training stages were 0.844, 3.76, and 0.964, 1.91, respectively. For the prediction of pyrolysis oil yield, the corresponding values were 0.926, 4.1, and 0.985, 1.889, in the testing and training stages, respectively.Temperature was identified as the most critical operating condition affecting pyrolysis oil and limonene yields, with relative importance percentages of 20.6% and 12.2% for oil and limonene yields, respectively. The optimal operating parameter range conducive to limonene yield is as follows: temperature between 350 and 450 °C, residence time between 0 and 50 min, and tire particle size greater than 15 mm.This study lays the foundation for the production of oil and the extraction of limonene from tire pyrolysis.