摘要
某些霍尔电压传感器在测试含有高频谐波的直流电压量时,存在着由纹波效应等原因引起的严重非线性误差问题,本文提出一种基于神经网络信息融合技术的传感器误差综合校正法。该法直接从传感器输出信号中提取纹波电压特征量,不需要附加检测传感器。将纹波电压作为非目标参量,输入电压作为目标参量,理想输出作为目标值,通过神经网络的训练后,获得了校正后网络的权值和阀值。仿真实验结果表明,采取从输出信号中提取融合信息的方式,利用神经网络的信息融合功能,能逼近一个校正平面,从而较好地解决了传感器误差综合校正问题。
Some of Hall voltage mutual inductors have serious error caused by ripple effect etc when DC voltage with HF harmonic is measured. In the face of this question, a sort of error comprehensive correcting method which is based on information fusion technology of the neural network is presented. On this way , the ripple voltage character parameter is picked-up directly from output signal of voltage sensor and the added sensor is not need. Regarding the ripple voltage as no-target parameter and input voltage as target parameter and ideal output as target value, an artificial neural network (ANN) is structured and trained. The weights and offsets of each layer are obtained after training the ANN. The simulating experimental results demonstrate that ANN can approach to a correcting plane and solve preferably the error - correction question of the Hall voltage sensor.
出处
《传感技术学报》
CAS
CSCD
北大核心
2008年第9期1571-1574,共4页
Chinese Journal of Sensors and Actuators
关键词
电气测量
综合校正
信息融合
BP神经网络
纹波效应
electric measurement
comprehensive correction
information fusion
BP neural network
ripple effect
作者简介
周克宁(1957-),男,分别于1982年和1998年在浙江大学获得学士和硕士学位,现为浙江科技学院副教授,主要研究方向为电气测控与电力电子技术,zkn60@yahoo.com.cn
徐然(1978-),男,工程师,1999年在浙江科技学院获学士学位,现为浙江科技学院工程师,主要研究方向为仪表技术及计算机控制系统技术,zust_xuran@126.com