基于RBF神经网络集成-模糊加权输出的数字温度传感器误差补偿  

Error compensation for digital temperature sensor based on RBF neural network ensembles-fuzzy weighing output

作　　者：林海军[1] 滕召胜[2] 杨进宝[1] 刘让周[2] 

机构地区：[1]湖南师范大学工学院,长沙410081 [2]湖南大学电气与信息工程学院,长沙410082

出　　处：《仪器仪表学报》2011年第7期1675-1680,共6页Chinese Journal of Scientific Instrument

基　　金：商务部优化机电和高新技术产品进出口结构(No.财企[2007]301号);湖南师范大学青年优秀人才培养计划(No.ET61107)资助项目

摘　　要：数字温度传感器存在非线性误差,在高精度测温系统中需要进行误差补偿。提出了一种基于径向基函数神经网络集成-模糊加权输出(RBFNNE-FWO)的数字温度传感器误差补偿方法:首先根据数字温度传感器的误差特征,提取特征阈值,构造三个相互独立的成员RBFNN;考虑到成员网络之间边界误差补偿问题,构建一种RBFNN集成输出权值模糊调节器,获得RBFNN集成输出权值,从而完成数字温度传感器的全量程误差补偿。与多种方法的比较仿真实验表明,这种RBFNNE-FWO方法的性能最佳、各成员网络边界误差最小,补偿后的数字温度传感器误差减少了两个数量级,大大提高了测温准确度。Nonlinear error compensation for digital temperature sensor is necessary in high accurate temperature measurement system.An error compensation method based on radial basis function neural network ensembles-fuzzy weighing output（RBFNNE-FWO） is proposed.The characteristic threshold values are obtained according the error characteristics of the digital temperature sensor and then three independent member RBFNNs are founded.Considering the compensation issue for the boundary errors among the member networks,a fuzzy weighing adjuster is established,and then the output weights of the RBFNN ensembles are obtained.Finally,the error compensation for the whole measuring range of the digital temperature sensor is realized.Simulation experiments were conducted and the proposed method was compared with different kinds of other methods;experiment results show that the performances of the RBFNNE-FWO method is the best,its boundary error is the smallest,and with the RBFNNE-FWO method the measuring error of digital temperature sensor decreases by two orders of magnitude.

关 键 词：数字温度传感器 误差补偿 径向基函数神经网络集成-模糊加权输出 边界误差 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置]