| At present,the crude oil distillation process generally has the problem of poor fractionation accuracy and high energy consumption.The main reason for this problem is that the coupling relationship between atmospheric distillation column and heat exchanger network is neglected,and only single unit optimization is considered The heat exchanger network is closely related to the distillation system and restricts each other.For example,when the heat exchanger network runs beyond the specified operating range,it will not only increase the energy consumption of the system,but also cause the internal heat imbalance of the distillation system,which directly affects the quality and yield of product.When the distribution of heat in the fractionator is not reasonable,the energy recovery rate of the heat exchanger network will be severely restricted.Therefore,the key to solve the above problems is to consider the optimization of cooperative operation between atmospheric distillation system and heat exchanger network,and to realize multi-objective optimization of profit and energy consumption of side-line products by adjusting the operating parameters of the equipment under the constraints of product quality.Due to the complexity of refinery production process,the structure of the mechanism model is very complex,and the optimization process needs to be evaluated by calling the model many times.Compared with each other,the direct adoption of the model will lead to serious computational complexity.It is difficult to meet the requirement of optimization efficiency.In addition,the data produced by the production process usually contain a lot of noise,and the data is seriously missing,so it is impossible to establish an accurate data model.The surrogate model assisted constrained optimization provides a new approach for solving the above optimization problems.The data model is used to replace the complicated mechanism model.According to the prediction accuracy of the surrogate model,the surrogate adaptively call mechanism model to update surrogate model,so that the surrogate can keep approaching the strict mechanism model.In recent years,more and more researchers use surrogate to solve expensive unconstrained optimization problems.The difficulty of solving this kind of problem lies in how to balance the constraint conditions and the objective function to select the update point of the agent model.Due to the fact that computationally time-consuming constraint optimization problems are common in practical applications,it is significant to constantly update and correct the surrogate by combining the constraint handling strategy with the model management strategy to solve the complex industrial optimization problems.In view of the above problems,this paper relies on the project of National Natural Science Foundation of China "Theory and implementation Technology of Global Cooperative Optimization of Refining production process(61590922)",and takes the cooperative operation optimization problem of fractionation system and heat exchanger network in refining process as the application background.The design and application of constrained multi-objective optimization algorithm based on surrogate model are studied.The main work of the study is as follows:(1)The mathematical description of cooperative operation optimization of distillation and heat exchange system in refining process is given.By analyzing the coupling relationship between the fractionator and the heat exchanger network,the decision variables,objective functions and constraints are determined.The decision variables include the feed temperature,thepump-around specifications,the flowrates of products and the injection amount of stripping steam.The constraint conditions include the model constraint between product profit,energy consumption cost and decision variables,the convergence constraint of simulation model operation and the constraint of product quality.The objective of optimization is to maximize net product benefits and minimize energy consumption costs.Under the constraint of product quality and other conditions,the optimization of key operating parameters can reduce energy consumption cost and improve product revenue,so as to realize the optimization integration between atmospheric distillation unit and heat exchanger network.(2)In order to solve the problem that the objective function and constraint conditions need long time to evaluate without convergence,a classification surrogate model based on fuzzy clustering under-sampling and adaptive Boosting is proposed to predict the convergent domain of atmospheric tower simulation model.It is a typical class imbalance problem to predict whether the atmospheric tower simulation model converges or not.The unbalance training data set is preprocessed by fuzzy clustering under-sampling method,and the individuals with the largest amount of information are uniformly selected from the categories with a large number of samples.The redundant samples are deleted,and the sample distribution is changed by using the above reconstruction sample training set.Adaptive Boosting integrated learning method establishes classification model to predict whether the input decision variables make the atmospheric tower simulation model converge so as to screen the non-convergent solution and improve the efficiency of performance evaluation.(3)In order to solve the problem that the feasible range is very small and disconnected,a constraint optimization evolutionary algorithm(K-CRVEA)based on surrogate model is proposed.Using gaussian process model as surrogate,an adaptive constraint management strategy based on the active state of reference vector is designed to update the surrogate.Firstly,according to the distribution characteristics of the current population,the reference vector is dynamically updated,and the individual is associated with the reference vector in order to ensure the diversity of the population and prevent the population from falling into local optimum.Then,in the early stages of evolution,according to the adaptive reference vector,the constraint value confidence lower bound criterion,the expected criterion and the feasible probability criterion are selected to determine the individual with good feasibility and high uncertainty.In the late stage of evolution,convergence and diversity are used as selection targets,and individuals with small constraint violation and good objective function are selected as the update points of the agent model.The experimental results show that the proposed algorithm can make the surrogate approach the feasible region quickly and find the global optimal solution quickly.(4)The design algorithm is applied to solve the optimization problem of distillation and heat exchanger system in refinery process.Firstly,the simulation model of atmospheric distillation unit is established by ASPEN HYSYS process simulation software,and the heat exchanger network model is established by pinch analysis technology based on thermodynamics principle,and the interface toolbox between MATLAB and HYSYS is compiled to ensure real-time communication between the two device models.The proposed algorithm is used to solve the cooperative operation optimization model,and is compared with the optimization algorithms EI-PoF and RVEA.The experimental results show that the proposed algorithm is effective in solving the coordinated operation optimization problem of distillation and heat exchanger network. |
PDF Full Text Request