Energy

Use case: Predict crude blend compatibility leveraging machine learning model

Context

Crudes blending crucial process in refining

  • To be able to extract value from non conventional crudes with high margins
  • But without compromising possible asset processing issues

Objective: Predict crude compatibility to optimize blend mix

Approach

Collected key available data

  • Compatibility lab analysis
  • Daily crudes blend
  • Crude cut points and assay

Selected key parameters for prediction and calculated "Blend Assay" for compatibility prediction Developed single regression analysis on crude assay characteristics to understand: correlations between variables


Tested 4 different models to estimate compatibility index but given the non linearity/multi factorial problem machine learning was needed.

Impact

Predictive dashboard to estimate compatibility index on different crudes blending with following features

  • crude assay integration
  • blending characteristic calculation
  • blending TBP curve
  • compatibility index

Prediction model accuracy: 80% on critical blends