RBF-LVLS方法用于甲醇合成反应器的软测量建模

RBF-LVLS Approach and its Application to Methanol Soft Sensor Modeling of Synthesis Reactor

及时、准确地测定化工过程变量,对确保生产过程稳定、有效控制产品质量具有重要意义。RBF-LVLS是在分析RBF-PLS的基础上提出的新方法,它保留了RBF-PLS的优点,采用非线性的神经网络结构,又用数学方法直接求解,免去了ANN冗长的训练过程和其他诸多欠缺,同时,它所集成的LVLS方法将PLS的多个目标函数整合为因变量成分拟合误差一个,以此循环迭代求解自变量和因变量的成分及它们间的回归系数,从而使建立的模型既具有很高的预报精度和良好的稳定性,又有简洁的解析形式,便于优化等进一步的计算和处理。RBF-LVLS方法成功应用于甲醇合成反应器的软测量建模。

It was well known that to measure and estimate the chemical process variables in time had vital significance in ensuring process stabilization and effectively controlling its product quality. In this study, the RBF-LVLS approach was proposed by analyzing the RBF-PLS method. The approach had the merit of RBF-PLS, i.e. using a structure similar to that of neural network, getting solution by mathematical methods directly, without the tedious training process of neural network and other evoking shortcomings. At the same time, the embedded latent variable least squares （LVLS） algorithm in the RBF-LVLS regression framework converted several objective functions into a function of the response variable fitting errors, and the PLS components and their regression coefficients between each pair of them were calculated based on this function. Thus, the RBF-LVLS could improve the accuracy and stability of predicted value of model. Moreover, the models had a brief analysis formula that was convenient for further processing such as optimization, Finally, it was successfully applied to soft sensor modeling of the methanol synthesis reactor.