Modular And Distributed Genetic Algorithm In Optimization Of Distillation

Distillation is still one of the most important operations in process engineering, and the application of mathematical programming tools to improve the efficiency of it has been vastly studied in the past years. Most of the researchers focused their attention on the generation of superstructures of distillation sequences and other simplifying methods to solve the synthesis of distillation systems. The formulation of such models, however, is difficult and time consuming. In addition, to compensate the difficulties in solving the very large non-convex models, it is often necessary to supply initial values for the optimization variables very close to the actual solution, something that is not always an easy task, and even recent works have used simplifications for the design model, thermodynamics, hydraulics, or the cost functions to obtain feasible solutions, which may pose negative effects on the accuracy of the final results. An alternative to reduce some of these problems is the use of genetic algorithm coupling with commercial simulators. But, because of the applying of rigorous models, the computational requirements are too high and become a major issue in extensive situations.A parallel genetic algorithm framework with integration of the commercial simulator ASPEN PLUS~@ is presented in this work. The use of modular simulators enables the process modeling to be accurate and convergent without the necessity of simplifications, while parallel computing environment guarantees the time efficiency of the optimization approach.At first, the genetic algorithm adopted here is discussed, including suitably designing of chromosomes and specialized genetic operators. To improve the performance of GA, several strategies like mixed code and random parent-number fitness-weighted crossover are implemented. Through the application in finding the optimal design of a heat-integrated methanol distillation system, we found that the traditional algorithm was not the most successful optimization algorithm on ourspecific domain. Then, a multi-step evolution method is included to the GA to compensate for the low speed based on law analysis of feedback control principle. The penalty function method applied to handle constraints is explained afterwards. The modular parallel genetic algorithm is realized by employing a network of PC with MPI message passing library as the parallel computing resources and by developing a interface in Visual Basic? 6.0 environment to control ASPEN Plus', version 11.1, used as simulator to evaluate the individuals generated by the parallel genetic algorithm. By the same time, probable genetic parameters are suggested too. A case is studied to illustrate the performance of the presented optimization framework and the employed problem is a novel process of three-column distillation proposed for the purification of synthetic methanol. Proven by the numerical results that the optimized distillation system can save 13.6% of the total cost annually when comparing with the original one, the proposed optimizing framework is succeeding in finding the optimal solution of a non-ideal heat-integrated distillation system. To evaluate the performance of multi-step evolution method, a comparison is made between a traditional GA implementation and the proposed implementation. It confirmed that the solution of the proposed algorithm is 26.9% better than that of the traditional one concerning about the annual costs. However, results generated by the GA with suggested size of population 50 after five runs clearly show the stochastic nature of genetic algorithms. The final solutions found in the five attempts even show a spread of anywhere up to 47%, which may be too big. To deal with that, we increase the size to 100, which proved by later results to be better in reducing the deviation to below 8%. Moreover, evaluation of time consuming in sequential and parallel cases proves that the presented algorithm is definitely time-efficient.