Machine Learning Predictions of Oil Yields Obtained by Plastic Pyrolysis and Application to Thermodynamic Analysis

Area of Science:

• Chemical Engineering
• Materials Science
• Machine Learning

Background:

• Chemical recycling via pyrolysis offers a route to convert plastic waste into valuable fuels and chemicals.
• Experimental yield determination for diverse plastic waste is often time-consuming and costly.
• Plastic waste composition significantly impacts pyrolysis product yields, especially for polymers like PET and PVC.

Purpose of the Study:

• To develop accurate predictive models for plastic pyrolysis oil yields based on feed composition.
• To assess the feasibility of using machine learning to guide plastic waste stream prioritization and pre-separation strategies.
• To evaluate the thermodynamic viability of pyrolysis for real-world plastic waste scenarios.

Main Methods:

• Compiled a dataset of 325 plastic pyrolysis experiments from open literature.
• Trained and evaluated seven machine learning regression models, including XGBoost.
• Validated model performance using a separate test dataset to determine accuracy (MAE).
• Applied the optimized XGBoost model to predict yields for simulated Municipal Recycling Facility (MRF) and Rhine River plastic waste compositions.

Main Results:

• eXtreme Gradient Boosting (XGBoost) achieved the highest prediction accuracy for oil yield, with a Mean Absolute Error (MAE) of 9.1% on the test set.
• The model successfully predicted oil yields for complex, real-world plastic waste mixtures.
• Analysis indicated that removing PET and PVC could be a viable pre-separation strategy to enhance oil yields.
• Thermodynamic analysis suggested that pyrolysis of Rhine River plastics is likely to be exergy-producing.

Conclusions:

• Machine learning, specifically XGBoost, provides a powerful and accurate tool for predicting plastic pyrolysis yields.
• Predictive modeling can significantly aid in optimizing plastic waste management and chemical recycling processes.
• The study demonstrates the potential for efficient energy recovery from diverse plastic waste streams through optimized pyrolysis.