基于LS-SVM的机械式温度仪表误差预测研究

机械式温度仪表在测量过程中易受到环境温度、毛细管长度以及内部机构影响而出现测量精度不高、非线性的情况,针对这些问题以液体压力式温度仪表作为研究对象,提出了基于最小二乘支持向量机(LS-SVM)的温度误差建模预测的方法。通过分析液体压力式温度仪表的测温结构和误差影响因素,将环境温度及毛细管长度等特征参数作为模型输入,将误差值及误差随毛细管长度的变化率作为输出。根据回归预测的原理,利用网格搜索和交叉验证的方法寻找最优参数组合,建立液体压力式温度仪表的误差预测模型。实验结果表明,该模型可以有效地描述温度误差,并将此建模方法与常用的支持向量机回归建模方法进行比较,基于LS-SVM得到的误差预测模型精度较高、推广能力强,可以对机械式温度仪表进行补偿,为探索机械式温度仪表自适应补偿机构提供理论依据。

Study on the temperature error prediction of mechanical temperature instrument based on LS-SVM

Aiming at the problems of poor accuracy and nonlinearity of mechanical temperature instruments caused by the influences of environmental temperature, capillary length and internal structure factors in the temperature measurement process, a new temperature error modeling prediction method for liquid pressure type thermometer based on least squares support vector machines (LS-SVM) is proposed to realize the temperature error compensation. Through analyzing the temperature measurement structure and error influencing factors of liquid pressure type thermometer, the environmental temperature and other structure parameters such as the capillary length and etc. are taken as the inputs of the model; the temperature error and the changing rate of temperature error vs capillary length are taken as the outputs of the model. According to the regression prediction principle, the grid search and cross validation methods are used to search the optimal parameter combination and build the error prediction model of liquid pressure type thermometer. Experiment results show that the proposed model can effectively describe the temperature error. The proposed modeling method was compared with conventional SVM regression modeling method; the result indicates that the error prediction model based on LS-SVM can achieve higher accuracy and stronger generalization capability. This method can be applied to the compensation of the liquid pressure type thermometer, and provides a theoretical basis for exploring the self-adaptive compensation mechanism used for mechanical temperature instruments.