稀疏LSSVM在4-CBA软测量建模中的应用  

Application of Sparse LSSVM in Soft Sensor Modeling of 4- CBA

作　　者：戎舟[1] 李佳庆[1] 

机构地区：[1]南京邮电大学自动化学院,江苏南京210003

出　　处：《仪表技术与传感器》2015年第12期88-91,共4页Instrument Technique and Sensor

基　　金：国家自然科学基金资助项目(61203213;11202107)

摘　　要：针对最小二乘支持向量机(LSSVM)缺失稀疏性问题,采用遗传算法对其模型进行稀疏化。算法思想如下:对LSSVM初始模型的核函数项进行二进制编码,采用遗传算法对二进制串进行寻优,将求得的最优个体解码,"1"代表选取该位置对应样本,"0"代表舍去该位置对应的样本,解码求得的样本集再次建模,重复上述稀疏过程,以每次测试样本相对误差的标准差为依据,当偏差率超过10%,则不再稀疏。将该算法应用于4-CBA(4-羟基苯甲醛)软测量建模过程,结果表明,采用遗传算法进行稀疏化的LSSVM模型,支持向量能稀疏70%左右,在保证预测精度的同时,大大提升了模型的效率。For the least squares support vector machine（ LSSVM） missing sparsity problem,the genetic algorithm（ GA） was used for sparse model. Idea was as follows： use binary coding method to code the kernals of initial LSSVM model. Then,GA was used to screen the binary strings. Decode the best individual. ＂1＂represents selecting corresponding position＇s sample and ＂ 0＂ represents truncating. Model again by the new sample. Repeat the above process. The algorithm is based on the standard of testing sample＇s relative error. When deviation rate of that is more than 10%,sparse operations end. The algorithm can be applied in soft sensor modeling of 4- CBA. The actual application result indicates that the sparse rate of support vectors of LSSVM model can reach about 70 percents. The algorithm improves the efficiency of model greatly without lowering the prediction precision.

关 键 词：LSSVM 稀疏化 遗传算法 软测量 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]