Simulation Based Approach to Optimal Design of Dividing Wall Column Using Random Search Method

Simulation based approach to optimal Design of dividing wall column using random search method

 

X. L. Ge,  X. G. Yuan (Corresponding author: yuanxg@tju.edu.cn)

State Key Laboratory of Chemical Engineering, School of Chemical Engineering and Technology, Tianjin University, Tianjin 300072, China

 

 

ABSTRACT

 

Optimal design of dividing wall column (DWC) involves multivariable optimization, which includes both continuous and discrete variables. And these variables interact with each other and need to be optimized simultaneously. Because of the more freedoms in terms of the numbers of stages as well as decision parameters of DWC comparing to conventional simple columns, the optimization problem turns out to be in large scale and combinatorial, and can be expressed as a mixed-integer nonlinear programming (MINLP) problem. However, the cost estimation and the material and energy balances of DWC are complex, and as a result, the objective function and constraints of the optimization problem are difficult to be formulated explicitly and directly as functions of the mixed variables. Thus, the optimal design of DWC is usually based on the detailed analysis, rather than a systematic approach.

In this work, a systematic optimization method was developed for optimal design of DWCs. The proposed method is based on the combination of a radial basis function neural network (RBF-NN) and genetic algorithms (GAs). The RBF-NN was built (trained) by a series of rigorous simulations of DWC for different sets of stage numbers with available tool provided in commercial software, ASPEN PLUS for example. In each of the simulations, the utility cost of the DWC for a specific set of stage numbers was optimized by adjusting the split ratios of respectively the liquid and vapor side draw streams from the main column to the overhead and the bottom of the prefractionator. Then, a GA was applied to optimize the stage numbers of the DWC with an objective (cost) function evaluated with the results obtained by the RBF-NN. The proposed method was validated with optimal design of DWCs for IDEAL, BTE and EPB systems respectively. The results were shown encouraging compared with those found in the literature. As a random search based approach, the proposed method showed its strength in finding the optimal solution by just evaluating small portion of the possible combinations of the stage numbers, and thus can be known as a promising method for optimal design of DWC.

 Keywords

Distillation, Dividing wall column, optimal design, simulation, random search