摘要
针对氢回收膜分离过程中一些重要性能参数难以在线测量,提出一种基于PCA-LSSVM的软测量模型。首先,用PCA分析影响气体膜分离过程的多个非线性变量间的相关性,得到它们各自对目标变量的重要程度。然后,用网格搜索、交叉验证结合贝叶斯估计,得到LSSVM模型的2个最优参数值(gam为0.7,sig2为100)。最后,建立模型并对数据进行预测。仿真结果表明,4输入变量(渗透气压力、尾气侧压力、原料气氢浓度、一段膜流量)PCA-LSSVM模型的预测值,与实测的渗透气氢浓度、渗透气流量和尾气氢浓度符合较好,最大相对误差都在8%以内,而且模型的收敛速度不到1 s,可为气体膜分离过程重要性能参数的在线检测和其过程的优化控制提供指导。
To aimed at the difficulty of measuring the performance parameters of hydrogen recovery membrane separation process in real time, a soft measurement model based on PCA-LSSVM was proposed. Firstly, PCA was used to analyze the relevance of multiple nonlinear variables, and obtain their different important degrees for the target variable. Then, combined grid search and cross validation with Bayes estimation were used to obtain the optimal value of two important parameters (gam =0.7, sig2 =100) of LSSVM. The simulation results show that the prediction results of the model with four variables (permeate-side pressure, residue-side pressure, feed hydrogen concentration, feed gas flux) are in reasonable agreement with the measurement values. All the maximal relative errors are less than 8%, and the convergence rate of the model is less than 1 s. This study provides a guidance for the on-line detection of important performance parameters of the gas membrane separation process and its process optimal control.
出处
《高校化学工程学报》
EI
CAS
CSCD
北大核心
2013年第5期877-883,共7页
Journal of Chemical Engineering of Chinese Universities
基金
海南省自然科学基金(211012)
国家科技支撑计划课题(2012BAA10B03)
国家973项目(2009CB623404)
国家自然科学项目基金(20736003
21176135)
关键词
气体膜分离
主元分析
最小二乘支持向量机
软测量
gas membrane separation
principal components analysis
least square support vector machine
soft measurement