Multiobjective Stochastic Optimization of Dividing-wall Distillation Columns Using a Surrogate Model Based on Neural Networks
Chemical processes have a multiobjective nature, since normally there are several objectives in conflict with each other; which are also restrained to requirements, physical or economical limitations. Since in multiobjective problems, the objectives are in conflict with each other, a simple solution is not desirable or sometimes even not feasible. For this kind of problem, a set of optimal solutions that represents the best trade-off between these objectives is the goal. These optimal designs can be achieved by means of Pareto front, which is a set of optimal non-dominated solutions1. In this way, the Pareto front allows having not just one optimal solution, but a set of optimal solutions that represents the best compromise between the objectives in consideration. Evolutionary algorithms have been recognized to be well suited for multiobjective optimization, because of their capability to evolve a set of non-dominated solutions distributed along the Pareto front2. One of the most popular multiobjective optimization algorithms is the Non-dominated Sorting Genetic Algorithm, NSGA-II3, which is a very robust tool and it is easy to implement. However, the principal disadvantage of genetic algorithms, and its variants, is the large amount of computational time that is often required for multiobjective optimization of industrial operations4; this fact without considering if the evaluation of the objective function is computationally expensive. This has led to the development of new strategies or combination of strategies to reduce the required computational time; basically, these strategies are classified as those that modify the parameters of the algorithm, and those using surrogate models. In the first type, key operators of the evolutionary algorithm are modified in order to give less randomness to the selection and generation of the individuals; the idea behind is to incorporate information about the problem, that can help to improve the search process. In general, the use of modified opMultiobjective Stochastic Optimization of Dividing-wall Distillation Columns Using a Surrogate Model Based on Neural Networks
Paper
Full text
Multiobjective Stochastic Optimization of Dividing-wall Distillation Columns Using a Surrogate Model Based on Neural Networks
Semantic Scholar · Chemistry · 2016
Abstract
Chemical processes have a multiobjective nature, since normally there are several objectives in conflict with each other; which are also restrained to requirements, physical or economical limitations. Since in multiobjective problems, the objectives are in conflict with each other, a simple solution is not desirable or sometimes even not feasible. For this kind of problem, a set of optimal solutions that represents the best trade-off between these objectives is the goal. These optimal designs can be achieved by means of Pareto front, which is a set of optimal non-dominated solutions1. In this way, the Pareto front allows having not just one optimal solution, but a set of optimal solutions that represents the best compromise between the objectives in consideration. Evolutionary algorithms have been recognized to be well suited for multiobjective optimization, because of their capability to evolve a set of non-dominated solutions distributed along the Pareto front2. One of the most popular multiobjective optimization algorithms is the Non-dominated Sorting Genetic Algorithm, NSGA-II3, which is a very robust tool and it is easy to implement. However, the principal disadvantage of genetic algorithms, and its variants, is the large amount of computational time that is often required for multiobjective optimization of industrial operations4; this fact without considering if the evaluation of the objective function is computationally expensive. This has led to the development of new strategies or combination of strategies to reduce the required computational time; basically, these strategies are classified as those that modify the parameters of the algorithm, and those using surrogate models. In the first type, key operators of the evolutionary algorithm are modified in order to give less randomness to the selection and generation of the individuals; the idea behind is to incorporate information about the problem, that can help to improve the search process. In general, the use of modified opMultiobjective Stochastic Optimization of Dividing-wall Distillation Columns Using a Surrogate Model Based on Neural Networks
References (63)
Scroll for more · 38 remaining