CC-95-126 OCTANE PREDICTION OF GASOLINE BLENDS USING NEURAL NETS

Description:

Blend modeling is usually approached with regression techniques applied to either bench blending or historical data from actual blends. Model maintenance is required as changes in products or components occur but is rarely performed. This paper discusses an approach to blend modeling using neural net technology to develop and maintain blend models for prediction of product octanes. The data collection, validation, and preparation for modeling as well as various aspects of the network structure are discussed.