Improving Group Contribution Methods by Distance Weighting

Abstract

A novel approach for improving estimations of existing groupcontribution methods is developed. Instead of fixed contributions,weighted contributions are optimized for each property estimation usinga database. These weighting factors are calculated based on similaritiesbetween the compound whose properties are estimated and the othercompounds in the database. By this approach, those components which arechemically more similar to the estimated one can systematically be givenmore weight in the estimation process, while sustaining the generalnature of the group contribution methods.The new approach was applied to the Joback and Reid (1987), and theMarrero and Gani (2001) group contribution methods. The performanceswere demonstrated on Normal Boiling Point (NBP) predictions. Theabsolute average error of NBP estimations was reduced by 62 K for theJoback and Reid method and by 4 K for the Marrero and Gani method. Otherphysical properties such as critical pressure, volume, formationenergies, melting temperature and fusion enthalpy were estimated byapplying the new technique to Joback and Reid method.