Towards time-effective optimization: Enviro-economic study of the C3MR LNG process

•Computation time shortened 7-fold by using Parallel Genetic Algorithm Interface.•18 process parameters studied and optimized in a ±75% interval.•Achieved 14% decrease in carbon dioxide emissions.•Liquefied natural gas processing costs simultaneously lowered by 12%. Liquefied natural gas (LNG) is an important energy carrier and one of the pillars of secure energy supply in many countries. However, energy requirements and the associated carbon footprint of its production are substantial. This, incombination with the rapidly growing production capacities, calls for efficient process design and operation. Arobustmulti-objective optimization study of a3.5 MTPA propane-precooled mixed-refrigerant (C3MR) LNG plant has been done employing the genetic algorithm (GA/NSGA-II), with 18 process parameters varying in a ±75% interval. The study has been performed in a novel Aspen Plus – Matlab interface for parallel flowsheet simulations which exploits the full computational capacity of a desktop computer. An over seven-fold increase in calculation speed has been achieved, allowing for a significant increase in the GA/NSGA-II population count (up to 1000 individuals) without the computations being excessively time-consuming. To choose the most suitable individual from the 1000 individuals in the final Pareto front, four decision making methods: Euclidean distance, fuzzy non-dimensionalized distance, and two statistical methods have been used and the results have been compared and discussed. As a result, amore than 12% reduction in LNG processing costs and a 14% cut in carbon dioxide emissions was achieved. This roughly translates into approx. 76 mil. USD/year decrease in the total annual processing costs and an over 76 KTPA decrease in the carbon dioxide emissions. In addition, the in-optimization behavior of process parameters was studied. Out of the 18 studied parameters, only six exhibited significant changes throughout theoptimization process while the others converged to a seeming overall optimum already in the first two runs. This implies a possibility of using similar parameter analyses in future optimization studies. Finally, results of the single- and dual-objective optimization were confronted with similar studies in the field and compared between each other. As the single-objective optimization yielded only marginal decrease in the respective objective function while significantly increasing the other one, the dual-objective optimization is favored.