Inventory Pinch Based Algorithm for Gasoline Blend Planning with Uncertainty in Components Qualities

In order to meet gasoline specifications, qualities of the blend components are analysed either in a lab or via on-line analysers. Deterministic gasoline blend planning assumes a quality of each blend component, even though there is an uncertainty associated with these qualities. In this work we present a chance constrained based methodology which enables a blend planner to decide what percent of the blends will meet specifications even if it means that some blends will give away qualities. This enables the planner to trade-off the cost of reblending the off-spec blends vs. the increase in the blend costs which guarantees that the quality specifications will be met. We employ a chance constrained formulation in (i) a two-level inventory-pinch based algorithm to optimize gasoline blend planning with uncertainty in components qualities; (ii) full space model for both linear and nonlinear blending rules. Inventory pinch based algorithm performs much better than the full-space formulation in terms of time required to reach the optimal solution, time to close the optimality gap to the required level, and robustness in achieving the optimal solution. Proposed algorithm enables construction of a trade-off curve of probability that the constraints will be satisfied vs. blend cost, which is not practical via full space formulation since the latter typically does not close the optimality gap or often it does not converge.