Intelligent Modeling And Optimization For The Molecular Weight Distribution Of Polyethylene

With good physical properties, chemical stability and processing performance, polyethylene is widely used in the industrial production and daily life. In polyethylene production process, the molecular weight distribution(MWD) is closely related with properties of products and is the most important indicator of quality for products. Therefore, the accurately modeling and real-time optimization for MWD of polyethylene have a very important practical significance.Due to the complexity and diversity of polymerization mechanism and the polydispersity of MWD, the traditional model of MWD with complicate, large-scale, complex structure, in practical application is limited. For the lack of effective modeling tool and real-time optimization method to MWD at this time, on the basis of previous theoretical studies, this paper has carried out some relevant research. The main research and innovation are as follows:(1)In polymerization process, the weighted superposition of distribution function on each active site of catalyst is used to fit the MWD of polyethylene to get a parameterized representation for MWD. By nonlinear least squares algorithm, the number of active site of catalyst, the weight and parameters of distribution function on each active site of the catalyst can be estimated. With the distribution function, the weight of distribution function and the model between operating conditions and parameters of distribution function, the model between operating conditions and MWD can be presented. As the model between operating conditions and parameters of distribution function is a multi-output modeling problem, the single-output algorithm is difficult to achieve the accurate model, the multi-output support vector machine regression algorithm(MSVR) is used to establish the model.(2) By the characteristics of selecting the optimal solution based on the fitness in single goal particle swarm optimization algorithm(PSO), in multi-objective PSO algorithm, the filter function is constructed based on the sum of percentage of the value on each dimension and the optimal solution on corresponding dimension of particle, combined with the congestion degree ordering policy and the roulette algorithm, the global optimal solution containing the most information of optimal solution can be selected. The simulation has shown this innovation could improve the quality of Pareto solution set. Finally, the improved multi-objective particle swarm optimization algorithm(MOPSO-CDRF) is applied to the ethylene polymerization process, based on the established model of MWD, to achieve the optimization of operating conditions.