基于SVR的传感器Hammerstein模型辨识

提出一种基于支持向量回归机的非线性动态传感器Hammerstein模型辨识方法并给出了相关的数学理论及学习算法.在该模型中,用非线性静态子环节和线性动态子环节串联来描述传感器的非线性动态特性.再利用函数展开将模型的非线性传递函数转换为等价的线性中间模型,并通过SVR求取中间模型参数.最后,推导出中间模型参数与传感器Hammerstein模型参数之间的关系,并由该关系实现非线性静态环节和线性动态环节的同时辨识.用实际力传感器动态标定实验数据进行测试,结果表明与常规非线性传感器辨识方法不同,所提方法只需进行一次动态标定实验就能给出非线性动态模型的数学解析表达式.且建立的力传感器Hammerstein模型阶次为4,而线性动态系统模型则需要6阶才能达到相同的精度.因此该研究为传感器非线性动态系统辨识又提供了一种可选方法.

Identification for Hammerstein Model of Transducer Based on Support Vector Regression

An identification method based on support vector regression(SVR)for Hammerstein model was investigated to analyze the nonlinear dynamic system of transducer.The corresponding mathematical theory and learning algorithm were also addressed.In this model,the nonlinear dynamic characteristic of transducer was expressed by a nonlinear static subunit followed by a linear dynamic subunit.Then,by the function expansion,the nonlinear transfer function of this model was converted to intermediate model which is linear one and can be identified using support vector regression(SVR).Finally,the relations of the coefficients of intermediate model and that of transducer's Hammerstein model were derived,through which the nonlinear static subunit and linear dynamic subunit were identified simultaneously.Practical dynamic calibrating experimental data of pressure-transducer were used to test.The results show that,compared with conventional identification methods,the proposed one can derive the analytic expressions of nonlinear dynamic transducer and only one dynamic calibrating experiment was needed.Furthermore,the order of Hammerstein model built by this method is 4,while,in the same precision,that of the linear one is 6.It provides a better way for identification for nonlinear dynamic transducer system.