基于ELM的非线性PLS软测量方法研究  

Study on Nonlinear PLS Soft Sensor Method Based on ELM

作　　者：毛志忠[1] 唐奡[1] 肖冬[1] 

机构地区：[1]东北大学信息科学与工程学院,辽宁沈阳110004

出　　处：《计量学报》2010年第4期374-378,共5页Acta Metrologica Sinica

基　　金：基金项目：国家自然科学基金（60674063）;国家“863”高技术研究发展计划（2007AA042194）

摘　　要：提出了一种基于极限学习机的非线性偏最小二乘软测量方法，有效地处理过程非线性、数据共线性以及在线快速建模等问题。该方法保留线性偏最小二乘外部模型以去除变量间的相关性，采用极限学习机建立内部模型来准确描述系统的非线性关系，与其它方法相比，该方法建模速度快，并具有良好的泛化能力。通过在苯乙烯反应器仿真及穿孔能耗预测应用结果表明，该方法建立的模型精度、泛化能力均优于传统方法，验证了其可行性和有效性。Based on extreme learning maehine,a nonlinear partial least squares soft sensor method is proposed to solve the problem of the process nonlinearity, data sharing, and especially the on-line rapid modeling in complicated industrial processes. The PLS outer projection is employed to remove the correlation between the variables, while the inter model with ELM precisely captures the nonlinearity. Compare with the traditional methods, the modeling time is much less and generalization performance is better in this soft sensor method. Applying this method to styrene reactor simulation and puehing energy consumption, the results proved its validity and feasibility.

关 键 词：计量学 非线性偏最小二乘 极限学习机 软测量 

分 类 号：TB9[一般工业技术—计量学]