A Scenario-Based Chance-Constrained Program for GasolineBlending under Uncertainty

Gasoline blending under uncertainty in the refinery valuechain optimization has gained tremendous attention. This paper proposes adata-driven chance-constrained programming approach to address this issueand guarantee the benefit of the refinery value chain. First, the blendingeffect model is introduced to capture the uncertainties in componentproperties, where the blending effect value is estimated from historicalprocess data by the recursive least-squares (RLS). Second, a chance-constrained gasoline blending model is proposed to ensure the on-specification products with a high probability in uncertain environments.Third, the Wasserstein generative adversarial networks (WGANs) areemployed to generate blending effect data unsupervised. Fourth, a scenario-based approach is used to reformulate the chance-constrained gasolineblending problem based on sufficient generated data. Accounting for thecomplexity of the resulting large-scale optimization, a sequential algorithm is applied to reduce the computational cost. Finally, anindustrial case study of gasoline blending is presented to demonstrate its applicability