基于自适应自编码器假人力学响应降维和重构方法

doi:10.3969/j.issn.1674-8484.2024.03.006

摘要/Abstract

摘要：

为了解决碰撞假人力学响应曲线的降维和重构问题，基于标准自编码器原理和假人力学响应曲线特征，添加限制条件，构建了自适应自编码器方法；选取假人的头部重心合成加速度曲线数据，经过数据清洗和取样后，作为样本数据；计算标准自编码器和自适应自编码器的互相关系数和重构均方误差，对比验证该自编码器的线性和非线性降维和重构能力。结果表明：该自编码器对假人力学响应数据的线性的升维重构误差为2.6%；对非线性的升维重构误差为2.4%；低维数据的协方差值接近于0。因此，本文所提方法可对假人力学响应数据实施线性和非线性降维，并可实现升维重构，且降维得到的低维数据具有强独立性。

关键词: 汽车安全, 汽车碰撞试验, 假人力学响应, 降维方法, 自编码器

Abstract:

A adaptive autoencoder was proposed to solve the problem of reducing and reconstructing the biomechanics response curve of collision dummy. The adaptive autoencoder method was constructed based on the standard autoencoder principle and the dummy biomechanics-response-curve characteristics with adding some constraints. The synthetic acceleration curve data were selected as sample data after data cleaning and sampling for the dummy-head gravity-center. The correlation number and the reconstruction mean square error of the standard autoencoder and the adaptive autoencoder are calculated; while the linear and nonlinear reduction and reconstruction ability were compared and verified for the adaptive autoencoder. The results show that the adaptive autoencoder has a reconstruction error of 2.6% for linear dimensionality and 2.4% for nonlinear dimensionality, while the covariance value is close to 0 for the low-dimensional data. Therefore, the adaptive autoencoder proposed in this paper implements linear and nonlinear dimensionality reduction and dimensionality reconstruction with a highly independent for the low-dimensional data.

Key words: vehicle safety, vehicle collision tests, dummy biomechanics response, dimensionality reduction technology, autoencoders

中图分类号:

U467

侯志平, 朱海涛, 刘灿灿, 杨佳璘. 基于自适应自编码器假人力学响应降维和重构方法[J]. 汽车安全与节能学报, 2024, 15(3): 337-343.

HOU Zhiping, ZHU Haitao, LIU Cancan, YANG Jialin. Dimensionality reduction and reconstruction method of dummy biomechanics response based on adaptive autoencode[J]. Journal of Automotive Safety and Energy, 2024, 15(3): 337-343.

图/表 10

参考文献 16

[1]	李海岩, 李广明, 贺丽娟, 等. 汽车追尾碰撞中颈部姿态对生物力学响应的影响[J]. 汽车安全与节能学报, 2022, 13(1): 55-62.
	LI Haiyan, LI Guangming, HE Lijuan, et al. Effect of neck posture on biomechanical response in rear end collision[J]. J Auto Safe Energy, 2022, 13(1): 55-62. (in Chinese)
[2]	陈现岭, 刘智超, 蒋斌庆, 等. 整车碰撞虚拟仿真中橡胶悬置的仿真方法[J]. 汽车安全与节能学报, 2020, 11(2): 212-219.
	CHEN Xianling, LIU Zhichao, JIANG Binqing, et al. Simulation method of rubber suspension in virtual simulation of vehicle collision[J]. J Auto Safe Energy, 2020, 11(2): 212-219. (in Chinese)
[3]	段利斌, 江浩斌, 张学荣, 等. 车身结构耐撞性与乘员保护虚拟仿真实验开发与实践[J]. 实验室研究与探索, 2020, 39(8): 120-124.
	DUAN Libin, JIANG Haobin, ZHANG Xuerong, et al. Development and practice of virtual simulation experiment of automotive body structure crashworthiness and occupant protection evaluation[J]. Res Expl Lab, 2020, 39(8): 120-124. (in Chinese)
[4]	逯小雨. 汽车安全碰撞仿真假人胸部结构设计研究[D]. 长春: 吉林大学, 2017.
	LU Xiaoyu. Structure design research of dummy's chest system for automobile safety impacting[D]. Changchun: Jiling University, 2017. (in Chinese)
[5]	王翔, 胡学钢. 高维小样本分类问题中特征选择研究综述[J]. 计算机应用, 2017, 37(9): 2433-2438+2448. doi: 10.11772/j.issn.1001-9081.2017.09.2433
	WANG Xiang, HU Xuegang. Overview on feature selection in high-dimensional and small-sample-size classification[J]. J Comp Appl, 2017, 37(9): 2433-2438+2448. (in Chinese)
[6]	XUE Bing, ZHANG Mengjie, Browne W N, et al. A survey on evolutionary computation approaches to feature selection[J]. IEEE Transa Evol Compu, 2015, 20(4): 606-626.
[7]	李晗, 刘钊, 朱平. 基于SSI-PSO的汽车碰撞试验时序数据处理与分类方法[J]. 汽车安全与节能学报, 2022, 10(2): 259-268.
	LI Han, LIU Zhao, ZHU Ping. Automobile crash test time-series data processing and classification method based on SSI-PSO algorithm[J]. J Auto Safe Energy, 2022, 10(2): 259-268. (in Chinese)
[8]	Sun W, Jin J H, Reed M P, et al. A method for developing biomechanical response corridors based on principal component analysis[J]. J Biomechanics, 2016, 49(14): 3208-3215.
[9]	Pulgar F J, Charte F, Rivera A J, et al. Choosing the proper autoencoder for feature fusion based on data complexity and classifiers: Analysis, tips and guidelines[J]. Info Fusion, 2020, 54: 44-60.
[10]	Portnova-fareeva A A, Rizzoglio F, Nisky I, et al. Linear and non-linear dimensionality-reduction techniques on full hand kinematics[J]. Front Bioengi Biotech, 2020, 8: 429-447.
[11]	苏琳. 结合知识图谱的变分自编码器零样本图像识别研究[D]. 阜新: 辽宁工程技术大学, 2023.
	SU Lin. Variational autoencoder combined with knowledge graph zero-shot learning[D]. Fuxin: Liaoning Technical University, 2023. (in Chinese)
[12]	Ranjan C. Understanding Deep Learning: Application in Rare Event Prediction[M]. USA: Connaissance Publishing Atlanta, 2020:281-329.
[13]	Alsenan S, Al-turaiki I, Hafez A. Autoencoder-based dimensionality reduction for QSAR modeling[C]// 2020 3rd Int’l Conf Compu Appl Info Security (ICCAIS). IEEE, Riyadh, Saudi Arabia. 2020: 1-4.
[14]	侯志平. 基于数据驱动的限制自编码器降维方法研究及其应用[D]. 天津: 河北工业大学, 2023.
	HOU Zhiping. Constraint autoencoder dimension reduction method via data driven and application[D]. Tianjin: Hebei University of Technology, 2023. (in Chinese)
[15]	HUANG Lei, LIU Xianglong, BO Lang, et al. Orthogonal weight normalization: solution to optimization over multiple dependent stiefel manifolds in deep neural networks[C]// Proc AAAI Conf Artif Intell, California, USA. 2018, 32(1): 3271-3278.
[16]	DUAN Shuyong, HOU Zhiping, HAN Xu, et al. A novel inverse procedure via creating tubenet with constraint autoencoder for feature-space dimension-reduction[J]. Int’l J Appl Mech, 2021, 13(8): 21500915-21500927

p	均方误差	低维数据间互相关系数
p	均方误差	最大值	最小值
50	0.032	5 152	0.180
40	0.050	4 209	0.004
30	0.130	2 401	0.030
20	0.350	1 402	0.580
10	1.480	2 004	7.780

p	均方误差	低维数据间互相关系数
p	均方误差	最大值	最小值
50	0.032	5 152	0.180
40	0.050	4 209	0.004
30	0.130	2 401	0.030
20	0.350	1 402	0.580
10	1.480	2 004	7.780

p	均方误差	低维数据间互相关系数
p	均方误差	最大值	最小值
50	0.026	0.035	0.000
40	0.130	0.071	0.000
30	0.440	0.099	0.000
20	0.930	0.135	0.000
10	1.750	0.017	0.000

p	均方误差	低维数据间互相关系数
p	均方误差	最大值	最小值
50	0.026	0.035	0.000
40	0.130	0.071	0.000
30	0.440	0.099	0.000
20	0.930	0.135	0.000
10	1.750	0.017	0.000

激活函数	标准自编码器均方误差	自适应自编码器均方误差
Leaky ReLU	0.016	0.024
Linear	0.032	0.026
SeLU	0.200	20.090
ReLU	0.750	16.280
Tanh	52.650	52.650