A Data-Efficient Framework for Automated Identification of Reaction Networks and Interpretable Rate Models

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Reaction Kinetics

Background:

• Mechanistic reaction rate models are crucial for reaction engineering but are difficult and time-consuming to develop.
• Current methods are expert-dependent and can introduce structural bias.
• Automated approaches are needed to improve efficiency and accuracy.

Purpose of the Study:

• To develop a two-stage automated framework for identifying minimal reaction networks and generating interpretable rate expressions.
• To enhance the accuracy, data efficiency, and physical consistency of reaction models.
• To accelerate knowledge discovery and enable the creation of physics-grounded digital twins in reaction engineering.

Main Methods:

• A two-stage framework combining sparse optimization for network identification and symbolic regression (SR) for rate expression generation.
• Introduction of a novel substructure-decomposition strategy to constrain SR for mechanistically meaningful and interpretable models.
• Evaluation on methanol synthesis and enzymatic reaction systems, incorporating model-based design of experiments.

Main Results:

• The framework accurately identifies minimal reaction networks and generates interpretable rate expressions.
• Demonstrated high accuracy and strong data efficiency across tested reaction systems.
• Ensured robust physical consistency of the developed models.

Conclusions:

• The proposed automated framework significantly accelerates the development of mechanistic reaction models.
• It overcomes limitations of traditional expert-driven approaches, reducing bias and labor.
• This work paves the way for enhanced knowledge discovery and the development of digital twins in reaction engineering.