A scalable optimization framework for refinery operation and management

Abstract

End-to-end refinery management is a complex scheduling problem requiring simultaneous optimization of coupled subprocesses at several stages. In the specific context of this paper, a planner needs to ascertain (i) how best to store incoming crude at a port, (ii) schedule its transfer, after dewatering, to downstream refinery tanks, and (iii) schedule further processing in the crude distillation units (CDUs). The movement and storage of crude is subjected to various physico-chemical and operational constraints. The resulting optimization problem is combinatorial in nature and scales exponentially with the number of tanks, types of crude, and modes of operation. The problem becomes particularly challenging with stochasticity in crude receipt, requiring the planner to modify their decisions in real-time. In this paper, we develop a scalable, hierarchical framework to address the end-to-end refinery management for throughput maximization. The framework relies on an innovative approach to decoupling the decision-making at port and refinery, reducing significantly the complexity of the overall optimization problem. The proposed approach also results in a significant improvement over the schedules generated by an expert human planner for throughput maximization. It takes only a few minutes to execute the entire optimization routine, over a 30 day planning window, on a standard computer, making it possible to use implement our approach in a time-critical, real-time operational setting.