| 采用BP神经网络和Burgers模型的细观参数标定 |
| |
| 引用本文： | 王洪波,马哲,乌兰图雅,樊志鹏,王春光.采用BP神经网络和Burgers模型的细观参数标定[J].农业工程学报,2022,38(23):152-161. |
| |
| 作者姓名： | 王洪波 马哲 乌兰图雅 樊志鹏 王春光 |
| |
| 作者单位： | 内蒙古农业大学机电工程学院，呼和浩特 010018 |
| |
| 基金项目： | 国家重点研发计划项目（2016YFD0701704-3）和内蒙古自治区自然科学基金项目（2020BS05022） |
| |
| 摘 要： | PFC软件作为一款成熟的离散元分析软件，由于在处理连续与非连续介质方面的出色表现，得到了广泛的应用。但是PFC软件所需要的细观参数均需要采用室内试验数据通过试错法反复调试才能获得，效率低、盲目性高，严重影响后续试验数据，因此急需一种新的细观参数校准方法。本文以玉米秸秆颗粒的单轴蠕变试验为基础，结合离散元软件PFC 2D，通过正交试验多因素方差分析方法分析了Burgers模型宏细观参数之间的影响关系，从而证明宏细观参数之间存在着复杂关系，不宜采用通过回归分析获得宏细观参数之间的关系式的方式标定细观参数，适合利用BP神经网络进行参数标定，利用创建的BP神经网络对细观参数进行标定，根据测试组的标定结果分析得出Burgers模型各细观参数的标定精度均在92%以上，且误差较为稳定，而且训练好的神经网络相关系数R>0.96，从而证明BP神经网络的细观参数标定性能较为可靠。将玉米秸秆单轴蠕变试验的宏观参数带入训练好的BP神经网络中进行细观参数标定，比对模拟蠕变试验与物理蠕变试验发现，两者的蠕变曲线基本一致，应变量的最大误差为2%，证明了BP神经网络具有良好的参数标定能力，可为PFC参数标定提供一定的参考价值。
|
| 关 键 词： | 离散元法 PFC 参数标定 BP神经网络 |
| 收稿时间： | 2022/9/2 0:00:00 |
| 修稿时间： | 2022/11/4 0:00:00 |
Calibration method of mesoscopic parameters using BP neural network and Burgers model |
| |
| Wang Hongbo,Ma Zhe,Wulantuy,Fan Zhipeng,Wang Chunguang.Calibration method of mesoscopic parameters using BP neural network and Burgers model[J].Transactions of the Chinese Society of Agricultural Engineering,2022,38(23):152-161. |
| |
| Authors: | Wang Hongbo Ma Zhe Wulantuy Fan Zhipeng Wang Chunguang |
| |
| Institution: | Mechanical and Electrical Engineering College of Inner Mongolia Agricultural University, Hohhot 010018, China |
| |
| Abstract: | As a mature discrete element analysis software, PFC software has been widely used because of its excellent performance in dealing with continuous and discontinuous media. However, the mesoscopic parameters required by PFC software can only be obtained by repeated debugging of laboratory test data through trial-and-error method, which has low efficiency and high blindness. Even experienced scholars need dozens of trial and error to obtain a set of usable parameters. For novice scholars, more trial and error are needed to obtain usable parameters. This matter requires scholars'' own parameter calibration experience, which seriously affects the promotion of PFC software and the follow-up test process. Therefore, there is an urgent need for a mesoscopic parameter calibration method that does not require the experience of scholars in parameter calibration, and can be correctly and quickly calibrated. Based on the uniaxial creep test of corn stalk particles and combined with the built-in Burgers model of the discrete element software PFC 2D, the uniaxial creep test model of corn stalk is established. The relationship between the macroscopic and mesoscopic parameters of the Burgers model is analyzed by the multivariate analysis of variance method of the orthogonal experiment, which proves that there is a complex relationship between the macroscopic and mesoscopic parameters. That is, the significance of the influence of each mesoscopic parameter on the macroscopic parameter is quite different, and the relationship between the macroscopic and mesoscopic parameters is highly nonlinear. Therefore, it is not appropriate to calibrate the mesoscopic parameters by obtaining the relationship between the macroscopic and mesoscopic parameters through regression analysis. However, these complex relationships are just suitable for parameter calibration using BP neural network. According to the number and characteristics of macroscopic and mesoscopic parameters, a BP neural network with 4 nodes in the input layer, 9 nodes in the hidden layer, and 5 nodes in the output layer is created through continuous attempts. Then, the created BP neural network is trained and calibrated using 150 sets of macroscopic and mesoscopic parameters. According to the results of the calibration of mesoscopic parameters using the created BP neural network, it is concluded that the calibration accuracy of all mesoscopic parameters in the Burgers model is above 92%, with relatively stable errors. Moreover, the correlation coefficient R of the trained BP neural network is greater than 0.96, which proves that the inversion performance of BP neural network is more reliable and it can be popularized and used as a new parameter calibration method for mesoscopic parameter calibration of Burgers model. Bringing the macroscopic parameters of the uniaxial creep test of corn stalk into the trained BP neural network for calibration of the mesoscopic parameters, and comparing the simulated creep test results with the physical creep test results, it is found that the creep curves of the two tests are basically the same, and the maximum error of the dependent variable is 2%, which proves that BP neural network has good parameter calibration ability and can provide certain reference value for PFC parameter calibration. |
| |
| Keywords: | DEM PFC parameter calibration BP neural network |
|
| 点击此处可从《农业工程学报》浏览原始摘要信息 |
| 点击此处可从《农业工程学报》下载免费的PDF全文 |
|
