PREDICTING TOP COMPOSITION OF TETRAHYFURAN USING XGBOOST

Overview

The case study is about predicting the top composition of the distillation column when the feed flow fluctuates. The distillation column separates THF and Toluene where the THF will be separated on the top while Toluene settles at the bottom. The feed is manipulated by adjusting the valve opening by ±5% to ±10%

Execution

The data is generated in Aspen HSYS and the parameters recorded are as shown below:

Where Master Comp Mole Fraction(TetraHyFuran) is the dependant variable. The XGBoost model were trained using 10-fold cross validation, with Mean Absolue Error as scoring metric

After assesing the model performance by feeding it test set, we can conclude the model perform very well in estimating the THF top composition

You can see how the model prediction mostly almost equal to the expected output: