Energy-integrated plastic-to-X optimization via machine learning-embedded graph-theoretic approach: EU 27 study

The growing environmental and economic challenges associated with plastic waste call for sustainable recycling solutions that align with circular economy principles and energy decarbonization goals. This work develops a novel machine learning-embedded multi-objective graph-theoretic (P-graph) optimization framework to evaluate plastic-to-X pathways across the EU-27, also considering Norway. Four recycling strategies are examined: mechanical recycling (plastic-to-plastic), pyrolysis to pyro-oil (plastic-to-fuel), gasification to methanol (plastic-to-chemical), and incineration (plastic-to-energy), each applied to four plastic compositions, i.e., mixed plastics (MP), polyethylene (PE), polypropylene (PP), and polystyrene (PS). The model integrates country-specific electricity mixes and normalizes performance based on both net total cost and CO 2 emissions using surrogate models derived from process simulations and literature data. Results show that pathway selection is influenced by both plastic composition and the national electricity profile. Mechanical recycling is economically favored for pure streams like PE, while pyrolysis dominates PS and PP due to high oil yields and low emissions. For MP, a hybrid approach combining mechanical recycling and methanol production is consistently selected. This work shows that cleaner energy mixes provide more options for plastic-to-X pathways when aiming to minimize costs and environmental impact at similar treatment potential. Further analysis of the top 1000 near-optimal solutions reveals that combining multiple pathways is attractive as opposed to single technologies. Overall, this work provides an ex-ante decision support framework for optimizing plastic waste valorization under region specific energy conditions.