改进型BP神经网络对电容称重传感器的非线性校正  

Nonlinear Calibration of Capacitance Weighing Sensor with Improved BP Neural Network Model

作　　者：郭伟[1] 张栋[1] 李巨韬[1] 王磊[1] 

机构地区：[1]天津大学机构理论与装备设计教育部重点实验室,天津300072

出　　处：《传感技术学报》2012年第10期1354-1360,共7页Chinese Journal of Sensors and Actuators

基　　金：青年科学基金项目(51005162);国家863计划项目(2011AA040601)

摘　　要：针对电容称重传感器电容检测电路的输出电压与载荷质量之间的非线性特性问题,基于贝叶斯正则化的L-M算法建立BP神经网络改进模型,实现了电容称重传感器的非线性特性的校正,并与传统的梯度下降算法建立的BP神经网络模型的校正结果进行了仿真对比。仿真结果表明:改进型BP神经网络具有较快的收敛速度、较高的精度和较好的推广能力,有利于准确实现电容称重传感器非线性特性的有效校正。Considering characteristics of the nonlinearity of the capacitance weighing sensor,i.e.the nonlinear relationship between the output voltage of the sensor and the loading,an improved BP neural network based on the Levenberg-Marquardt algorithm of Bayesian-Regularization was established to improve the nonlinear calibration capabilities.Simulation results show that the improved BP neural network achieved faster rate of convergence,higher accuracy and stronger generalization capability in comparison with the traditional gradient descent algorithm,which can effectively upgrade the nonlinear calibration of the capacitance weighing sensor.

关 键 词：电容称重传感器 非线性校正 贝叶斯正则化 LEVENBERG-MARQUARDT算法 梯度下降算法 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置]