Quantitative structure-property relationship studies for predicting flash points of alkanes using group bond...

Area of Science:

• Computational Chemistry
• Chemical Engineering
• Machine Learning

Background:

• Flash point is a critical safety parameter in chemical engineering.
• Predicting flash points from molecular structure is essential for chemical process design.
• Existing methods like multilinear regression have limitations in capturing complex structure-property relationships.

Purpose of the Study:

• To develop a predictive model for alkane flash points based on molecular structure.
• To utilize artificial neural networks (ANNs) for modeling structure-flash point relationships.
• To assess the accuracy and applicability of the developed ANN model.

Main Methods:

• Employed the group bond contribution method to define molecular structure descriptors.
• Utilized a back-propagation (BP) artificial neural network (ANN) to model nonlinear structure-property relationships.
• Trained and validated the ANN model using a dataset of 92 alkanes, with data split into training, validation, and testing sets.

Main Results:

• The optimized BP neural network ([9-5-1]) demonstrated high accuracy in predicting flash points.
• Achieved an average absolute deviation of 4.8K and a root mean square error (RMS) of 6.86.
• The ANN model outperformed multilinear regression in predictive accuracy.

Conclusions:

• The developed ANN model effectively correlates alkane molecular structure with flash point.
• The model provides a reliable tool for predicting alkane flash points in chemical engineering.
• This approach enhances the understanding of quantitative structure-property relationships for alkanes.