(689d) Global Optimization for Refinery Operations Planning Using Complex Unit Models

Refinery operations require careful planning, which is handled these days by different MINLP formulations. Existing Commercial software  does not guarantee global optimality. Incentives to search for global optima are reflected in the fact that just 1% improvement in the gross refinery margin represents sizable profit. In addition, we recognize that several units in the refinery are reactors, which, if modeled in detail, are represented by differential equations. Therefore, any improvement in the direction of more rigorous modeling requires the incorporation of these differential equations in the planning model. Instead of attempting the explicit incorporation of these differential equations, we resort to creating input-output models, which are used in a global optimization method based on domain and image partitioning. Domain partitioning takes place the usual way. Image partitioning simply creates images of the discrete partitioning points. Once this is performed, a lower bound model and a recently developed bound contraction model is used to obtain the global solution.