基于SVM的传感器动态模型辩识方法  

Dynamic Model Identification Method Based on SVM for Sensors

作　　者：吴德会[1,2] 王晓红[1] 

机构地区：[1]九江学院电子工程学院 [2]合肥工业大学仪器仪表学院,合肥230009

出　　处：《传感技术学报》2006年第3期716-719,共4页Chinese Journal of Sensors and Actuators

摘　　要：提出了一种传感器动态模型辩识新方法,给出了相应的辩识过程及学习算法。该方法采用支持向量机模型,与常规模型辩识方法比较,其优点是明显的。其采用了结构风险最小化准则,在最小化样本误差的同时减小模型泛化误差的上界,提高了模型的泛化能力;而且将学习算法转换为求解二次规划问题,使得整个模型参数辩识过程中有且仅有一个全局极值点,确定了结果的唯一性。最后,仿真和实际试验结果均表明应用支持向量机对传感器动态模型进行辩识有效。This paper presents a new method for dynamic model identification of sensors, and gives the corresponding design processes and learning algorithm. This method is based on support vector machine （SVM） model. Compared with conventional model identification methods, this method possesses prominent advantages：over fitting is unlikely to occur by employing structural risk minimization criterion to minimize the errors at the samples and decrease simultaneously the upper bound of the predict error of the models. Furthermore, the global optimal solution can be uniquely obtained owing to that learning algorithm converts machine learning into quadratic programming. Simulations and experimental results show the dynamic model identification method is not only effective but of high precision.

关 键 词：支持向量机 动态模型 传感器 模型辩识 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置]