Accurate Prediction of Polymerization Performance for Metallocene Catalysts via a Dual-Path Neural Network and Local...

Area of Science:

• Polymer Chemistry
• Catalysis
• Materials Science

Background:

• Metallocene catalysts are crucial for producing tailored polypropylene via propylene homopolymerization.
• Rational catalyst design is hindered by complex structure-condition relationships and limitations of current machine learning (ML) models.
• Existing ML approaches often neglect key ligand descriptors, impacting their industrial applicability.

Purpose of the Study:

• To develop a hybrid ML framework integrating reaction parameters and catalyst structural features for improved polypropylene synthesis.
• To accurately predict catalyst activity and polymer properties, overcoming limitations of conventional methods.
• To provide a computational tool for targeted catalyst design and scalable material production.

Main Methods:

• A dual-path neural network was employed to process numerical and categorical inputs separately, preventing feature semantic distortion.
• A k-nearest neighbor regression model was utilized to capture local sample relationships for predicting narrow molecular weight distributions.
• The hybrid framework integrated catalyst structural features and reaction parameters.

Main Results:

• The dual-path neural network achieved high accuracy in predicting catalyst activity (R² = 0.9201) and number-average molecular weight (R² = 0.9133).
• The k-nearest neighbor model demonstrated superior performance (R² = 0.9766) for predicting narrow molecular weight distributions in polypropylene.
• Both developed models outperformed eight other benchmark ML algorithms.

Conclusions:

• The hybrid ML framework offers a robust and interpretable computational strategy for linking catalyst chemistry to polymer properties.
• This approach facilitates the targeted design and scalable application of high-performance polypropylene materials.
• The study provides a practical tool for advancing metallocene catalyst development.