Symbolic Machine Learning Enhances Vapor-Liquid Equilibrium Predictions for Hydrocarbon-Nitrogen Mixtures
A new arXiv preprint (2608.11255) introduces a symbolic machine learning approach to improve vapor-liquid equilibrium (VLE) predictions for hydrocarbon-nitrogen binary mixtures. The method corrects the Peng-Robinson equation of state (PR-EOS) using interpretable symbolic expressions derived from experimental data. The two-level strategy first identifies symbolic corrections for individual hydrocarbon systems, then models the coefficients as functions of carbon number to generalize across different hydrocarbons. Results show significantly improved accuracy over the original PR-EOS for all tested systems. The work addresses the challenge of accurate VLE prediction across broad composition ranges and hydrocarbon chain lengths, offering a balance between the interpretability of classical equations of state and the accuracy of deep learning models.
Key facts
- The paper is available on arXiv with ID 2608.11255.
- The approach uses symbolic machine learning to discover corrections to the Peng-Robinson equation of state.
- The method employs a two-level strategy: per-system symbolic expressions, then coefficients as functions of carbon number.
- The results show improved prediction accuracy over the original PR-EOS for hydrocarbon-nitrogen systems.
- The work targets VLE prediction for hydrocarbon-nitrogen binary mixtures.
- The approach aims to provide interpretable analytical expressions.
- The study covers a broad range of compositions and hydrocarbon chain lengths.
- The paper was announced as a new submission on arXiv.
Entities
Institutions
- arXiv