基于即时学习算法的软测量建模方法  

Soft-sensing Method Based on Lazy Learning Algorithm

作　　者：王其红[1] 潘天红[2] 邹云[3] 

机构地区：[1]常州信息职业技术学院自动控制工程系,江苏常州213164 [2]江苏大学电气信息工程学院,江苏镇江212013 [3]南京理工大学自动化学院,江苏南京210094

出　　处：《南京理工大学学报》2007年第6期679-683,722,共6页Journal of Nanjing University of Science and Technology

摘　　要：针对工业过程中存在许多难以直接测量变量的问题,提出1种基于即时学习算法的软测量建模方法。提出k-矢量近邻方法建立系统的当前工况点的建模邻域,以提高即时学习算法的预测能力,利用递归最小二乘算法简化建模的在线计算量,通过PRESS估算得到算法的最优解,优化了算法,应用该算法对某一化工厂产品的酯化率进行测量建模,预测得到的最大相对误差为0.8742%,证明该方法的泛化性能良好,满足实际测量的精度要求,是1种便于理解,易于实现的软测量方法。In view of lots of unmeasured variables in industrial process, a soft sensing method using lazy learning is presented. The k-vector nearest neighbor is used to generate a neighbor of current regime in order to enhance the predictive capability of the original algorithm. The complexity of the algorithm is decreased by recursive least squares and the optimal solution is addressed by PRESS. Using this method to model ester rate of a local chemical plant, this paper obtains the maximum inaccuracy of 0. 874 2%. The simulation results show that the perfect generalization performance can meet high-precision measuring requirements and the method is very understandable and easy to be implemented.

关 键 词：即时学习 k-矢量近邻 软测量 酯化率 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]