论文部分内容阅读
环氧乙烷浓度是乙二醇生产过程中的1个重要指标,其浓度大小直接影响到后续水合反应生成乙二醇的过程。环氧乙烷浓度与多种因素之间存在着复杂的非线性关系,在软测量建模的过程中消除这些因素的相关性可以有效地降低计算复杂度。本文综合应用主元分析法,粒子群优化算法以及径向基函数神经网络建立了环氧乙烷浓度的软测量模型。首先分析了影响环氧乙烷浓度的因子,并对这些因子进行了主成分分析,得到1组新的输入因子。然后按照累积方差贡献率选取合适的输入因子,作为RBF神经网络的新的输入,有效降低了输入变量的维数,减少了输入变量之间的相关性,简化了神经网络的结构,建立了环氧乙烷浓度的软测量模型。最后利用粒子群算法来优化神经网络参数,求解RBF网络的径向基中心和输出层连接权值的最优值,减少了计算时间,提高了计算精度,获得了较好的拟合和预测效果。与只采用RBF网络建立软测量模型相比,本文采用的方法建模的误差较小,计算时间较短,计算精度较高,网络的预测效果较好。
Ethylene oxide concentration is an important indicator of ethylene glycol production process, the concentration of a direct impact on the size of the subsequent hydration reaction of ethylene glycol process. There is a complicated non-linear relationship between the concentration of ethylene oxide and many factors. Eliminating the correlation of these factors in the process of soft-sensing modeling can effectively reduce the computational complexity. In this paper, the principal component analysis, particle swarm optimization algorithm and radial basis function neural network are used to establish the soft sensor model of ethylene oxide concentration. First of all, the factors that affect the concentration of ethylene oxide were analyzed, and the principal component analysis of these factors was carried out to obtain a new set of input factors. Then according to the contribution rate of cumulative variance, we select the appropriate input factor as the new input of RBF neural network, which effectively reduces the dimensionality of input variables, reduces the correlation between input variables, simplifies the structure of neural network, Soft-sensing model for the concentration of oxygen ethane. Finally, particle swarm optimization algorithm is used to optimize the parameters of neural network to solve the RBF RBF center and output layer connection weights of the optimal value, reducing the computational time and improve the accuracy of the obtained better fitting and prediction results . Compared with using only the RBF network to establish the soft sensor model, the method proposed in this paper has less error, shorter computation time and higher computational accuracy, and the network has a better prediction effect.