Refinery Solutions > Gasoline Blending > Chemometric Modeling - Best Practices

Chemometric Modeling - Best Practices

  • Chemometric modeling package – Solo software from Eigenvector Research, Inc.
  • Model types available
    • PLS1
    • PCA2
    • PCR3
  • Conservative choice of factors from SECV4 table
  • Outliers from leverage vs. studentized residuals
  • Normalization method depends upon sample type
  • Cross-validation options
    • Venetian blinds
    • Only one pass-through
  • Generalized least squares weighting (GLSW) filtering
    • Reduces subtle differences between instruments and process streams
1PLS = partial least squares regression;
2PCA = principal components analysis;
3PCR = principal components regression;
4SECV = standard error of cross-validation 

GLSW filtering greatly reduces the number of factors
Improves model performance and transference
Hierarchical Models

  • Normally, one model is sufficient to cover the seasonal blend changes observed in most properties
  • Solo + Process Instruments PROspect real-time software
    • Create hierarchical models that can easily be tailored to smaller seasonal ranges      
      • Improves standard error of prediction (SEP) for critical parameters
        • Octane
        • Reid vapor pressure (RVP)
        • Distillation points
        • More
  • Process Instruments PROspect automatically runs a particular parameter through the correct target model, depending on its prediction via a broader, trigger model
  • Modeling is included for 12 months following installation
    • During this period, we can train your personnel to handle modeling
  • After 12 months, we can continue to handle modeling on a fee-for-service basis
    • Typical model turnaround is ~ 3 business days after receipt of data spreadsheet