摘要
针对车辆动态称重过程中称重信号受外界干扰导致称重精度不高的问题,提出结合互补集合经验模态(CEEMD)和门控循环单元(GRU)神经网络的称重算法。采用CEEMD对原始称重信号进行分解,得到的残余分量为初步滤除干扰信号的轴重信号;然后将残余分量归一化后作为GRU神经网络输入层构建网络模型,输出车辆轴重。研究结果表明,除去个别因异常数据导致的不良结果,该方法车辆动态称重误差控制在1.2%以内。相比于传统的单一模型,称重精度更高,实用性更强。
Aiming at the problem of low weighing accuracy caused by external interference in the weighing process of dynamic truck scale,a weighing algorithm combining complementary ensemble empirical mode decomposition(CEEMD)and GRU neural network was proposed.Firstly the original weighing signal was decomposed by CEEMD,and the residual component obtained was the axle load signal that preliminarily filtered out the interference signal.Then the residual component was used as the input layer of GRU neural network to construct the network model and output the vehicle axle load.The research results show that the vehicle dynamic weighing error of this method is controlled within 1.2%after removing individual abnormal data caused by measurement error.Compared with the traditional single model,it has higher weighing accuracy and stronger practicability.
作者
杨帮
赖征创
杨晓翔
YANG Bang;LAI Zhengchuang;YANG Xiaoxiang(College of Mechanical Engineering and Automation,Fuzhou University,Fuzhou 350108,China;Fujian Province Institute of Metrology,Fuzhou 350003,China;Quanzhou Normal University,Quanzhou 362000,China)
出处
《中国测试》
CAS
北大核心
2023年第9期108-114,共7页
China Measurement & Test
基金
福建省属公益类科研院所基本科研专项项目(2019R1016-2)
泉州市科技局科技计划项目(2020C055)。
关键词
动态称重
互补集合经验模态分解
门控循环单元神经网络
信号处理
深度学习
dynamic weighing
complementary ensemble empirical mode decomposition
gated recurrent unit neural network
signal processing
deep learning