Multi-Objective AI Optimization of Plastic Waste Pyrolysis Integrating Energy Return on Investment for Circular Polymer Recycling.

A rapid accumulation of plastic waste has created an urgent need for efficient and sustainable recycling technologies. Among various approaches, pyrolysis stands out as promising method of thermochemical recycling of plastic waste; however, the process needs optimization and further research to make it more energy-efficient and sustainable. The conventional approaches for optimization focus on the enhancement of yield, only overlooking efficiency and system-level sustainability. In this study, a machine learning-enabled surrogate-assisted multi-objective artificial intelligence (AI) optimization framework is developed for plastic pyrolysis to maximize product recovery and minimize energy consumption. The model integrates energy return on investment (EROI) and higher heating value (HHV) into process design. A curated dataset of 312 experimental cases covering polyolefins, PET, nylon, and mixed plastics was used to train multiple machine learning algorithms, such as polynomial regression, Gaussian process regression, and Random Forest models. The Random Forest algorithm demonstrated superior predictive robustness across oil yield, HHV, char formation, and EROI. Pareto front analysis using NSGA-II revealed that moderate reaction severities (400-450 °C, 40-70 min) maximize net energy performance while minimizing solid residues. The conditional variational autoencoder as a GenAI model was incorporated to work as a generative proposal engine, which enhances the exploration of chemically feasible operating regions under uncertainty-aware active learning. The integration of techno-economic and life-cycle assessment demonstrates that energy-positive configurations outperform high-yield scenarios, achieving IRR > 15%, energy intensity < 10 MJ kg