Kriging-assisted constrained optimization of single-mixed refrigerant natural gas liquefaction process

Framework for embedding black-box simulation into mathematical programming via kriging surrogate model applied to natural gas liquefaction process optimization

“…Nikkho et al [43] optimized two mini-scale modied 3-stage compression with phase separation, 2-stage expansion SMR natural gas liquefaction processes using a GA. Santos et al [44] investigated the optimization of a 4-stage compression with phase separation, 1-stage expansion SMR natural gas liquefaction process employing an augmented number of decision variables with Nelder-Mead derivative-free optimization method, considering valve and hydraulic turbine expansion. Later the methodology was improved in Santos et al [45], in which a kriging-assisted global search scheme that included the optimization of the probability of feasible improvement acquisition function to nd promising candidates to run the local search with Nelder-Mead algorithm.…”

“…Given the present review on methods for black-box optimization problems and single-mixed refrigerant natural gas liquefaction process design and considering the particularities of the simulation optimization problem from this design task toward minimum energy consumption using reliable chemical process simulators, surrogate modeling can be used to introduce symbolic formulation to the optimization problem functions that then can be embedded in mathematical programming setup and solved using classical and ecient gradient-based optimization or deterministic global optimization. Dierently from what was done in [39] and [45], which optimized the surrogate optimization problem or acquisition function based on the surrogate models using global optimization meta-heuristics, the present approach explore the mathematical information introduced by the surrogate models. It means that, generic regression models are tted to data generated from the rigorous simulation and used to replace the black-box functions f and g by surrogates f and ĝ ĝ ĝ that introduce analytical formulation to those functions with reliable derivatives that can be used for ecient gradient-based optimization of the resulted nonlinear programming problem.…”

“…For mathematical background on the surrogate model used in this surrogate-based optimization framework, the kriging model is derived as the best linear unbiased predictor, Gaussian process regression model following the derivation of Sacks et al [8], Lophaven et al [47], Stein [48], and Santos et al [45].…”

Structural Modifications on Single Mixed Refrigerant Cycles for Performance Improvements in Small-Scale LNG Processes

“…Santos et al investigated the optimization of the PRICO cycle to minimize its shaft power demand using surrogate models (known as Kriging) to replace the more computationally demanding rigorous models from Aspen HYSYS. Their work uses the particle swarm algorithm to explore the solution space and find a candidate that maximizes the probability of improving the objective function.…”

“…Also, the works from Khan et al and Xu et al provide an effective procedure to manipulate the composition of the mixed refrigerant (which is a key degree of freedom), although only applied to the PRICO cycle. In other works, − information about the optimal refrigerant composition obtained from pinch and exergy analysis is usually limited. These analyses rarely provide insights regarding adequate refrigerant compositions to minimize shaft power demand.…”