Multi-objective Evolutionary Optimization Of Chemical Process Based On Surrogate Model

The continuous advancement of society has led to an unprecedented consumption of energy in all walks of life,especially for petrochemical products.The operation optimization of petrochemical products involves many performance indicators,and each optimization target is clamped to each other.It is difficult to effectively improve the index parameters through traditional optimization methods,and it is time-consuming and laborious to optimize the operation of the actual model.Research on multi-objective optimization of chemical process based on agent model is very necessary.As a typical production unit in the chemical industry,the rectification unit effectively increases production capacity,reduces energy consumption,reduces pollution,and enhances economic benefits.It has become an important proposition for the national economy and the people's livelihood.According to this background,this thesis studies the multi-objective optimization of chemical rectification process based on agent model.The specific research contents are as follows:1.Facing the multi-dimensional high-dimensional nonlinear operating system such as the rectification process,simply constructing the mechanism model in the chemical process simulation platform such as Aspen Plus to optimize the processing not only brings great challenges in calculation,but also takes time and is extremely serious.Based on this,this thesis chooses the radial basis function RBF proxy model to replace the mechanism model.According to the different mechanism of the proxy model in the modeling process,the proxy model can be divided into offline proxy model and online proxy model.The offline agent model is often modeled by one-time adding points.There are many sample points required,the accuracy of the model cannot be automatically adjusted and affected by the spatial density and uniformity of the sample points.The online agent model can pass the small number of modeling sample points.Adaptive sampling to update the model until convergence,and can effectively use the sample point distribution information of the previous generation model to guide the subsequent modeling process.Therefore,this thesis selects the online agent model,and combines the previous point-adding strategy to propose adaptive sampling adjustment coefficients.It can continuously update the values according to the precision design indicators in the process of building the model,and select multiple comparisons from each generation of Pareto optimal solution sets.Excellent candidate sampling points are used to reconstruct the model to meet the accuracy requirements.Finally,the optimal search is developed based on this,and the performance of the algorithm is tested by using the classic Benchmark function.2.Because the operation optimization like rectification process is faced with multi-objective problems,in order to effectively weigh the contradictory relationship between multiple targets,this thesis adopts the fast non-dominated sorting genetic algorithm NSGA-II with elite strategy as multi-objective.The optimization algorithm performs an optimal search for the rectification process.Aiming at how to effectively combine proxy model and multi-objective optimization algorithm,this thesis proposes Adaptive Sampling RBF Surrogate Based Multi-objective Optimization(ASS-MOP).The multi-objective optimization algorithm NSGA-II can effectively utilize part of the modeled sample points.With each iteration,the candidate sample points are reasonably selected from the Pareto surface and added to the sample set according to the change of the sampling point threshold.Therefore,the accuracy of the model is gradually improved,and finally the According to this model,the Pareto optimal solution is obtained.3.Apply the ASS-MOP algorithm to the rectification equipment.By comparing the ASS-RBF Model,the Pure surrogate model and the Rigorous Model based on the adaptive sampling strategy,it is found that the ASS-RBF Model is compared to the pure mechanism model.The sampling time is greatly reduced.The Hypervolume(HV)value of the solution quality evaluation index is largely close to the real model,indicating that the Pareto optimal solution searched by the ASS-RBF Model is sufficient to replace the real optimal solution.Through the comparative analysis of the ASS-RBF Model's own modeling efficiency,it is found that the Pareto frontier surface of the model tends to be stable at the later stage,but the slight adjustment is made on the original basis,which shows that the ASS-RBF Model has relatively better optimization efficiency.