基于BP神经网络的天平校准数据处理方法研究  

Research on Data Processing Method of Balance Calibration Based on BP Neural Network

作　　者：车兵辉[1] 尹欣繁 彭先敏[1] 章贵川[1] Che Binghui;Yin Xinfan;Peng Xianmin;Zhang Guichuan(China Aerodynamics Research and Development Center,Mianyang 621000,China)

机构地区：[1]中国空气动力研究与发展中心,四川绵阳621000

出　　处：《计算机测量与控制》2020年第10期165-169,共5页Computer Measurement &Control

摘　　要：针对基于最小二乘法的多项式天平校准公式拟合方法无法消除天平非线性引起的误差,提出了基于BP神经网络的公式拟合方法;理论分析了典型的3层BP神经网络的结构和学习原理,介绍了常规多项式公式拟合和BP神经网络公式拟合步骤和方法,建立了具有6个输入节点和6个输出节点的三层BP神经网络,采用C语言实现了BP神经网络的训练过程和自动学习,并保存训练结果;采用某六分量天平校准数据,将天平输出电压值作为BP神经网络的输入,将天平加载载荷作为网络输出对网络进行训练,并给出了多项式和神经网络方法的计算误差对比结果;结果表明,采用神经网络方法拟合精度平均提高了67%,可有效消除系统非线性引起的误差,天平公式拟合精度显著提高。In view of the fact that the fitting method of the calibration formula of the polynomial balance based on the least square method cannot eliminate the error caused by the balance nonlinearity,a formula fitting method based on the BP neural network is proposed.The structure and learning principle of typical 3-layer BP neural network are analyzed theoretically,the steps and methods of conventional polynomial formula fitting and BP neural network formula fitting are introduced,and a three-layer BP neural network with 6 input nodes and 6 output nodes is established,the training process and automatic learning of BP neural network are realized by using C language,and save the training results.The calibration data of a six-component balance were used,the output voltage value of the balance was used as the input of BP neural network,the load of the balance was used as the output of the network to train the network,and the comparison results of the calculation error between the polynomial and the neural network method were given.The results show that the error caused by the system nonlinearity can be eliminated effectively by using the neural network method,and the fitting accuracy of the balance formula is improved six-seven percent.

关 键 词：BP神经网络 天平公式 拟合 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]