作者：史柏迪1,2, 庄曙东1,2, 陈威1, 陈天翔1, 朱楠楠1
作者单位：1. 河海大学机电工程学院，江苏 常州 213022;
2. 南京航空航天大学 江苏省精密与微细制造技术重点实验室，江苏 南京 210016
关键词：动态测量系统;极限梯度提升树;长短期记忆神经网络;岭回归;补偿系统
摘要：
物流秤在动态测量过程中，各类时序干扰信号极易对压力传感器测量精度造成影响。基于正交试验法获取物流秤在不同带速、载重及采样频率下台面的压力及秤体三轴加速度信号，以此作为样本集，基于五折交叉验证原则依次建立岭回归、Xgboost以及改进的LSTM测量补偿模型。结果表明Ridge模型具有最低的算法复杂度，且较传统线性回归模型提升明显，补偿平均损失为0.317 kg；Xgboost模型平均损失为0.219 kg且基于F检验分析误差成分；此外提出一种改进的LSTM神经网络模型，通过在原有结构基础上堆叠全连接层，将采样信号作为时间序列变量输入模型，最终测试结果表明虽模型训练时间与空间复杂度较大，但补偿测量准确度最佳损失低至0.142 kg，且对采样频率不敏感具有最好的鲁棒性。
Establishment of time series weighing model under dynamic factors
SHI Baidi1,2, ZHUANG Shudong1,2, CHEN Wei1, CHEN Tianxiang1, ZHU Nannan1
1. College of Mechanical and Electrical Engineering, HoHai University, Changzhou 213022, China;
2. Jiangsu Key Laboratory of Precision and Micro-manufacturing Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China
Abstract: In the dynamic measurement process of logistics scale, all kinds of time sequence interference signals are easy to affect the measurement accuracy of pressure sensor. Based on the orthogonal test method, the pressure signal of the platform and the three-axis acceleration signal of the scale body under different belt speed, load and sampling frequency are obtained. Based on the 50 fold cross validation principle, Ridge regression, xgboost and LSTM measurement compensation models are established in turn. The results show that ridge model has the lowest algorithm complexity, and the average compensation loss is 0.317 kg compared with the traditional linear regression model; the average loss of xgboost model is 0.219 kg, and the error components are analyzed based on F-test; in addition, an improved LSTM neural network model is proposed, which directly takes the sampled signal as time series by stacking the full connection layer on the basis of the original structure The final test results show that although the training time and space complexity of the model is large, the best compensation measurement accuracy loss is as low as 0.142 kg, and it is insensitive to the sampling frequency and has the best robustness.
Keywords: dynamic measurement system;Xgboost;LSTM;Ridge regression;compensation system
2021, 47(7):135-141 收稿日期: 2020-07-10;收到修改稿日期: 2020-08-20
基金项目: 江苏省高校实验室研究会立项资助研究课题（GS2019YB18）；江苏省精密与微细制造技术重点实验室数学建模课题组(CZ520007812)；中央高校基本科研业务费（2018B44614）
作者简介: 史柏迪(1996-),男,江苏常州市人,硕士研究生,专业方向为寿命预测
参考文献
[1] 许白玲. 互联网金融治理体系与优化策略[J]. 技术经济与管理研究, 2020, 3(4): 91-95
[2] 白莲. 线上线下购物中的消费者关注差异研究[J]. 商业经济研究, 2019, 4(15): 88-91
[3] LIU Y K , YU X, XUE J P, et al. A flexible phase error compensation method based on probability distribution functions in phase measuring profilometry[J]. Optics and Laser Technology, 2020, 9(4): 45-51
[4] 郑大宇, 孟庆鑫, 王立权, 等. 带传动动力学分析及惯性力影响的研究[J]. 哈尔滨工程大学学报, 2008, 13(9): 73-77
[5] 董祥辰. 基于FPGA的皮带动态秤动态称重神经网络算法的实现[D]. 南京: 南京理工大学, 2018.
[6] PIETRZAK P, MELLER M, NIEDZWIECKI M. Dynamic mass measurement in checkweighers using a discrete time-variant low-pass filter[J]. Mechanical Systems and Signal Processing, 2014, 48(12): 67-76
[7] YE W, ZHANG Y. CEO traits, dynamic compensation and capital structure.[J]. PloS one, 2019, 14(2): 22-31
[8] IRWAN S, TSUYOSHI U W. In-ear microphone measures in the ear canal with bone conduction stimulation: An application for estimating a cross-talk compensation filter[J]. Acoustical Science and Technology, 2020, 41(3): 115-121
[9] HUSSAIN M, REHAN M, AHMED S, et al. A novel approach for static anti-windup compensation of one-sided Lipschitz systems under input saturation[J]. Applied Mathematics and Computation, 2020, 3(7): 65-73
[10] 赖朝晖, 杨晓翔, 姚进辉, 等. 汽车衡安装误差对称重精度的影响分析[J]. 中国测试, 2019, 45(7): 108-116
[11] 竺元昊,范伟军,郭斌, 等. 汽车气压盘式制动器性能在线检测系统研制[J]. 中国测试, 2020, 46(10): 138-143
[12] 杨楠. 岭回归分析在解决多重共线性问题中的独特作用[J]. 统计与决策, 2004(3): 14-15
[13] 张剑. 动态称重数据处理算法及其在禽蛋和类球形水果分选中的应用研究[D]. 杭州: 浙江大学, 2017.
[14] HE K M, ZHANG X Y, REN S Q, et all. Delving deep into rectifiers: surpassing human-level performance on imagenet classification[J]. Computer Science, 2015, 47(12): 131-141
[15] 周志华. 机器学习[M].北京: 清华大学出版社, 2018.
