Modeling and optimization of liquefied natural gas process

Abstract

Energy is a global concern. Although natural gas, the cleanest fossil fuel, is a fast growing primary energy source for the world today, its transportation is a critical technological issue. While liquefied natural gas (LNG) offers the most economic way of transporting natural gas over long distances, most LNG plants are energy-intensive, use large amount of utilities and are often designed and operated based on enumerative and heuristics based approaches. In contrast, this PhD work demonstrates the application of advanced modeling and optimization techniques with significant improvement in overall energy efficiency and costs for existing LNG plants. Specific focus is given to operational modeling, heat and fuel integration, and global optimization.

First, a novel approach for deriving an approximate operational model for the complex and proprietary multi-stream heat exchanger (MSHE) that produces LNG in an LNG plant is presented to predict its performance over a variety of seasons and feed conditions without knowing its physical details but using operational data only. Second, the traditional heat exchanger network synthesis (HENS) is extended to accommodate non-isothermal phase changes which are abound in MSHEs, condensers and reboilers in LNG plants. The proposed generalized HENS or GHENS model includes non-isothermal phase changes of process and utility streams, allows condensation and/or evaporation of mixtures, permits streams to transit through multiple states and improves the annualized cost of heat integration significantly. Third, using the concept of source-sink superstructure, a mixed-integer nonlinear programming (MINLP) model is developed for optimizing the operation of fuel gas networks and a case study from an existing LNG plant is presented. This successfully integrates fuel sources such as boil-off gases produced in various parts of an LNG plant and demonstrates significant savings in operating and energy costs. Finally, since model nonlinearities and nonconvexities often prevent commercial solvers to obtain global solutions, the global optimization of nonconvex bilinear problems is addressed. Focus is given to the development of piecewise linear relaxation of nonconvex bilinear terms, for which a bivariate partitioning scheme is presented. Such relaxation is shown to provide better lower bounds when solving bilinear programs (BLP) and mixed integer bilinear programs (MIBLP) to optimality.