一种基于LS-SVM构造FLANN的热电偶非线性校正方法  

Rectification of Non-Linearity for Thermocouple Based on FLANN Constructed by LS-SVM

作　　者：吴德会[1,2] 王晓红[1] 

机构地区：[1]九江学院电子工程系 [2]2.合肥工业大学仪器科学与光电工程学院,合肥230009

出　　处：《传感技术学报》2007年第6期1321-1324,共4页Chinese Journal of Sensors and Actuators

基　　金：国家自然科学基金资助(70272032)

摘　　要：提出一种基于最小二乘支持向量机(LS-SVM)构造函数链接型神经网络(FLANN)的方法,并根据正反馈原理将该FLANN应用於热电偶传感器非线性校正.讨论LS-SVM构造FLANN的基本原理和具体算法,给出了非线性补偿器的数学模型.与常规BP迭代算法构造的FLANN比较,该方法构造的FLANN补偿器具有如下优点:①利用LS-SVM将迭代逼近问题转化为直接求解多元线性方程,因此具有更快的速度;②整个训练过程中有且仅有一个全局极值点,确定了所构造FLANN补偿器的唯一性,提高了补偿精度.最后以Pt-Rh30-Pt-Rh6热电偶(B型)为例进行非线性校正实验,结果验证了上述结论.A least squares-support vector machine （LS-SVM）-based method was proposed to construct functional link artificial neural networks （FLANN）, which was used to correct non-linearity of thermocouple sensor based on the principle of positive feedback. The principle and algorithms of LS-SVM-based FLANN were discussed and the non-linearity compensation model for thermocouple was given. Compared with tra- ditional BP-based FLANN, the new LS-SVM-based FLANN had more advantages： ①the LS-SVM solution solved a set of linear equations instead of an iterative problem, so it was faster in speed; ② FLANN, which was uniquely obtained due to the global maximum in the whole training process, was higher in accuracy. Finally, Platinum-Rhodium 30-Platinum-Rhodium 6 thermocouple （B） sensor was taken as an example to non-linearity compensation, and experimental results show the presented method is effective.

关 键 词：最小二乘支持向量机 函数链接型神经网络 热电偶传感器 非线性校正 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置]