首页|基于GRU-BP算法的高精度动态物流称重系统

基于GRU-BP算法的高精度动态物流称重系统

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
针对动态物流秤测量精度对载重、采样频率、带速较为敏感的问题,提出了一种高精度动态物流称重系统.首先,采用三因素五水平正交试验法,结合皮尔逊相关性检验原则,使用低通巴特沃斯与卡尔曼滤波器对传感器压力信号进行了滤波降噪处理,并将加速度信号作为模型输入信号,进行了特征补偿;然后,基于深度学习算法,提出了一种改进的门控循环单元模型,在该模型采样区间内将压力与振动改写为时序化信号,并将其共同输入门控循环单元(GRU)模型;最后,对GRU模型进行了改进,对其结构输出了层堆叠误差反向传播神经网络(BP),有效加强了模型的非线性映射能力.研究结果表明:在各类传动速度及测试货物下,该模型的最大测量误差相对于同类型深度学习模型长短期记忆(LSTM)神经网络、循环神经网络(RNN)时序模型及传统数值平均模型的误差,依次降低了16.14%、27.14%、76%,可用于各类称重系统.
High-precision dynamic logistics weighing system based on GRU-BP algorithm
Aiming at the problem that the measuring accuracy of dynamic logistics scale was sensitive to load,sampling frequency and belt speed,a high-precision dynamic logistics weighing system was studied.Firstly,using three-factor and five-level orthogonal experimental method,combined with the Pearson correlation test principle,the low-pass Butterworth and Kalman filters were used for filtering and noise reduction of the sensor pressure signals,and the acceleration signals were used as the input signals of the model for the feature compensation.Then,an improved gated recurrent unit model was proposed based on a deep learning algorithm,in which the pressure and vibration temporalized signals in the sampling interval were jointly input into the gated recurrent unit(GRU)model.Finally,the GRU model was improved,and the nonlinear mapping ability of the model was effectively enhanced through the back-propagation(BP)neural network of stacking errors in the output layer.The research results show that the maximum measurement error of this model can be respectively reduced by 16.14%,27.14%and 76%compared with that of the same type of deep learning model,such as long short term memory(LSTM)neural network,recurrent neural network(RNN)timing model and traditional numerical average model,under various transmission speeds and test goods.It can be used in all kinds of weighing systems.

deep learningdynamic measurement systemgate control loop unit(GRU)back-propagation(BP)neural networkvibration compensationlong short term memory(LSTM)neural networkrecurrent neural network(RNN)

康杰

展开 >

三江学院 机械与电气工程学院,江苏 南京 210012

深度学习 动态测量系统 门控循环单元 反向传播神经网络 振动补偿 长短期记忆神经网络 循环神经网络

江苏省高等学校自然科学研究面上项目教育部产学合作协同育人项目

19KJD510005201902168015

2024

10.3969/j.issn.1001-4551.2024.06.020

机电工程
浙江大学　浙江省机电集团有限公司

机电工程

CSTPCD北大核心
影响因子：0.785
ISSN：1001-4551
年,卷(期)：2024.41(6)