Uncertain Distillation Process Optimization By Chance Constrained Programming With Recourse

The uncertain variables in the process can make the existing design and operating parameters are not the best choices and even cause the process unreasonably. The traditional processing method leads to the decision nonoptimal and even the whole process is in potential safety problems. So it is necessary to research the optimization problems with uncertainty. Due to the known optimization models and the solution methods can not optimize the process with several uncertain parameters well. Therefore it is very necessary to hunt for more reasonable optimization models. In this paper we consider the uncertain variables in the optimization process respectively. Establishing the penalty function of some uncertain parameters, and stochastic programming with recourse is used to research these uncertain parameters. The rest of uncertain parameters are considered by chance constrained programming. So that the chance constrained programming with recourse model is established. And it can compensate for the defects of using any kind of stochastic programming model, such as it is difficult to find the penalty function of some uncertain variables and solve the stochastic programming with recourse and chance constrained programming with one or more uncertain variables. The new model can be used to ensure the reliability of the process. It also can be good to restrain the difficulty of solving the problem.The computational method of the stochastic programming with recourse and chance constrained programming strategy is to change the problem to the equivalent deterministic problem. Then use LP or NLP to solve the deterministic problem. The uncertain variables are in the stochastic programming with recourse and chance constrained programming respectively. So using sequence reforming process to transform the new model. At first, change chance constraint into the equivalent nonlinear constraint. Because the traditional method of conversion chance constrained is feasible only on special occasions. Sampling methods need large amount of calculation. It is difficult to change the chance constrained by numerical integral, because it needs multiple integral. So use Newton-Cotes to get the determined nonlinear constraint. And then use a hybrid algorithm of both Monte Carlo integration and improved Benders decomposition strategies to translate the stochastic programming with recourse to the deterministic model through active constraint to restrict the infeasible pionts. It reduces the computational load. At the last use SQP to solve the deterministic problem. Use the stochastic programming model which is put forward in this paper to optimize distillation process. Study uncertainty optimization of the simulated experiment of methanol and water separation process and the separation process of ethylene propylene generating 1-hexene. The product market demand, feed rate, feed composition and the plate efficiency are regarded as uncertain parameters in methanol and water separation process. The optimization results show the income of process at high confidence level is lower than that of process at short confidence level, and the income of process when the confidence level is 0.5 is near to the deterministic optimization result. Feed composition, the steam price of low pressure and high pressure are uncertain variables in the separation process of 1-hexene. We can get some optimization results of three separation schemes. The cost of scheme 1 is the lowest when the confidence level is 0.5 and 0.6, but the cost of scheme 3 is the lowest when the confidence level is no less than 0.65. The scheme 1 is the best from the deterministic optimization result as the result of low confidence level, So that the selected scheme and decision variables will lose its significance in the actual process.Through the research of distillation process we can know the optimization model which we set up in the paper can provide reliable reference value to determine decision variables of the process. And it also can raise the reliability of the decision variables and provide guidance to the future uncertain optimization research.