Page 352 - 《振动工程学报》2025年第11期
P. 352
第 38 卷第 11 期 振 动 工 程 学 报 Vol. 38 No. 11
2025 年 11 月 Journal of Vibration Engineering Nov. 2025
融 合 特 征 降 维 与 人 工 神 经 网 络 的 坐 姿 人 体 头 部
振 动 特 性 研 究
张筱璐, 陈相玉, 段远飞, 孙浩宇, 林 森
(北京工业大学机械与能源工程学院,北京 100124)
摘要:从座椅传递至人体头部的低频振动会影响乘员的舒适性,甚至会导致晕动症等疾病。本文基于低频人体振动试验,构
建一种用于预测从座椅到头部传递函数的人工神经网络模型;并针对体征参数间的耦合相关导致的模型多重共线性问题,分
别采用主成分分析法和核主成分分析法对模型的体征参数特征进行降维优化,以准确预测不同乘坐环境下的座椅到头部传
递函数。结果显示,在不同靠背接触条件下,经主成分分析法降维后的优化模型与反向传播人工神经网络模型相比,在预测
单轴或多轴激励下座椅到头部传递函数时,预测精度均显著提升。在预测不同坐姿条件下(有、无靠背接触)座椅到头部传递
函数时,融合核主成分分析法相较于融合主成分分析法的模型,预测误差分别进一步降低了 2.5% 和 28.7%,预测精度分别达
到 0.9657 和 0.9676。研究表明,对于不同坐姿条件下的多轴振动激励环境,核主成分分析法比主成分分析法能够更有效地减
少模型参数的冗余信息,并提高对座椅到头部传递函数的预测精度。
关键词: 人体振动;座椅到头部传递函数;体征参数;人工神经网络;特征降维
中图分类号:TP393 文献标志码:A DOI:10.16385/j.cnki.issn.1004-4523.202508081
Modeling of the seat-to-head transmissibility based on the artificial neural network
and the feature dimensionality reduction method
ZHANG Xiaolu,CHEN Xiangyu,DUAN Yuanfei,SUN Haoyu,LIN Sen
(College of Mechanical and Energy Engineering,Beijing University of Technology,Beijing 100124,China)
Abstract:Low-frequency vibrations transmitted from the seat to the head of the seated individuals can impact riding comfort and even induce
motion sickness. Based on the whole-body vibration experiments,artificial neural network models to predict the seat-to-head transmissibility
were developed. To address the multicollinearity issue caused by anthropometric parameters and to enable accurate prediction of the seat-to-
head transmissibility,the dimensionality of the anthropometric parameters in the model was optimized by employing both principal component
analysis and kernel principal component analysis. The results indicate that under conditions with and without a backrest, the prediction
accuracy of the seat-to-head transmissibility for single-axis or multi-axis excitation was significantly improved in the model incorporating
principal component analysis, compared to the backpropagation artificial neural network model. When predicting the seat-to-head
transmissibility under conditions with and without a backrest, the model fused with kernel principal component analysis exhibits further
reduced prediction errors by 2.5% and 28.7%,achieving accuracies of 0.9657 and 0.9676,respectively,compared to the principal component
analysis integrated model. The research suggested the kernel principal component analysis is more efficient than the principal component
analysis at minimizing redundant information,thereby enhancing the prediction precision of the seat-to-head transmissibility.
Keywords:whole-body vibration;seat-to-head transmissibility;anthropometric parameters;artificial neural network;feature reduction
车辆行驶等振动环境下的人体头部振动响应可 座椅到头部传递函数(seat-to-head transmissibility,
能会影响乘员的视觉功能和驾乘舒适性,甚至可能导 STHT)常用于量化从座椅至人体头部的振动传递特
致晕动症等疾病 [1-2] 。因此,构建用于预测振动环境下 性 。集中参数模型、多体动力学模型和有限元模
[3]
座椅到人体头部振动传递函数的模型,对优化座椅设 型已被广泛用于分析和预测 STHT 。其中,集中参
[4]
计及避免人体振动损伤具有重要的实践意义。 数模型由质量块、阻尼和弹簧组成,多用于分析振
收稿日期:2025-08-15;修订日期:2025-09-27
基金项目:国家自然科学基金资助项目(51605010)

