<p>The conventional method for acquiring vehicle load spectrum struggle to meet the rapid development needs of vehicles. Based on data samples constructed by virtual proving ground (VPG) technology, a fast prediction method for load spectrum with an long short-term memory (LSTM) surrogate model is proposed. Firstly, the vehicle model is updated and validated through suspension kinematics &amp; compliance (K&amp;C) testing and load spectrum calibration. Secondly, an updated vehicle model is used to construct load spectrum training dataset incorporating mass, stiffness, and road surface information. The trained LSTM model can achieve rapid prediction of load spectrum from road surface information. Furthermore, the influence of sampling frequency and road type on the accuracy of the load spectrum prediction is studied in detail. Finally, the proposed method was verified through the measured load spectra of SUV models developed on the same platform. The results demonstrate that the coefficient of determination (<i>R</i><sup>2</sup>) between the predicted loads and the measured data was verified to exceed 0.91, and the computational time required for the VPG method to generate a 10-second load spectrum was reduced from 30 minutes to less than 1 second using the surrogate model method. The results demonstrate that the LSTM surrogate model provides an efficient and reliable approach for platform-based vehicle development.</p>

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Research on rapid prediction of vehicle load spectrum based on LSTM surrogate model

  • Chun Zhang,
  • Junru He,
  • Ruoqing Wan,
  • Jian Yu

摘要

The conventional method for acquiring vehicle load spectrum struggle to meet the rapid development needs of vehicles. Based on data samples constructed by virtual proving ground (VPG) technology, a fast prediction method for load spectrum with an long short-term memory (LSTM) surrogate model is proposed. Firstly, the vehicle model is updated and validated through suspension kinematics & compliance (K&C) testing and load spectrum calibration. Secondly, an updated vehicle model is used to construct load spectrum training dataset incorporating mass, stiffness, and road surface information. The trained LSTM model can achieve rapid prediction of load spectrum from road surface information. Furthermore, the influence of sampling frequency and road type on the accuracy of the load spectrum prediction is studied in detail. Finally, the proposed method was verified through the measured load spectra of SUV models developed on the same platform. The results demonstrate that the coefficient of determination (R2) between the predicted loads and the measured data was verified to exceed 0.91, and the computational time required for the VPG method to generate a 10-second load spectrum was reduced from 30 minutes to less than 1 second using the surrogate model method. The results demonstrate that the LSTM surrogate model provides an efficient and reliable approach for platform-based vehicle development.